call for proposals - whut.edu.cnkfy.whut.edu.cn/zxtz/201507/p020150721397820585235.pdf · enabled...

1 / 206

Call for Proposals

HIRP OPEN 2015

2 / 206

Copyright © Huawei Technologies Co., Ltd. 2015. All rights reserved.

No part of this document may be reproduced or transmitted in any form or by any means without prior written consent of Huawei Technologies Co., Ltd.

Trademarks and Permissions

and other Huawei trademarks are trademarks of Huawei Technologies Co., Ltd.

All other trademarks and trade names mentioned in this document are the property of their respective holders.

Confidentiality

All information in this document (including, but not limited to interface protocols, parameters, flowchart and formula) is the confidential information of Huawei Technologies Co., Ltd and its affiliates. Any and all recipient shall keep this document in confidence with the same degree of care as used for its own confidential information and shall not publish or disclose wholly or in part to any other party without Huawei Technologies Co., Ltd’s prior written consent.

Notice

Unless otherwise agreed by Huawei Technologies Co., Ltd, all the information in this document is subject to change without notice. Every effort has been made in the preparation of this document to ensure accuracy of the contents, but all statements, information, and recommendations in this document do not constitute the warranty of any kind, express or implied.

Distribution

Without the written consent of Huawei Technologies Co., Ltd, this document

cannot be distributed except for the purpose of Huawei Innovation R&D

Projects and within those who have participated in Huawei Innovation R&D

Projects.

3 / 206

Wireless Communication Technology（HIRPO201501） ................................................................. 6

Subject: Visible light communication ........................................................................................ 6

Subject: Design Massive MIMO for Energy Efficiency ............................................................... 9

Subject: Reconstruction or approximation to FDD Downlink Channel Information ............... 11

Subject: High reliable wireless network technology for buildings .......................................... 13

Subject: Capacity Analysis of SCMA Systems .......................................................................... 16

Subject: Wireless signal based gesture recognition technology ............................................. 19

Future Networks（HIRPO201502） ............................................................................................... 21

Subject: Research on Internet of Things (IoT) carrier network architecture and technology . 21

Subject: Bearer Network for 5G .............................................................................................. 26

Subject: Research on in network computing .......................................................................... 28

Subject: Research on DCE NIC offloading ................................................................................ 30

Optical Technology（HIRPO201503）............................................................................................ 32

Subject: RF Crosstalk Analysis in CFP2-DCO module ............................................................... 32

Subject: All optical metro network and key technology collaboration ................................... 36

Storage Technology（HIRPO201504） ........................................................................................... 46

Subject: Research on New programming model based on NVM ............................................ 46

Subject: Storage optimization Research for NoSQL DBs ......................................................... 51

Subject: Research on Distributed Transaction Management Framework and Prototype ....... 56

Subject: Research on Data Placement Algorithm on Distributed Storage Scene .................... 61

Subject: Research on Extra Lower Latency Interconnection Network .................................... 69

Subject: Research on IO capture and analysis tools prototype based on the network side in

NAS system .............................................................................................................................. 73

Subject: Research on 3D TLC NAND Flash Reliability .............................................................. 76

Subject: Research on SCM based Data mirroring and remote replication .............................. 79

Subject: LUT-based low resource LDPC solution ..................................................................... 81

Subject: The Power Loss Character of Nand when programming and data recovery ............. 83

Subject: ReRAM switching mechanism study and MLC/TLC feasibility ................................... 84

Subject: UFS hardware acceleration performance improvement solution ............................. 85

Subject: UFS firmware architecture solution .......................................................................... 87

Computing Technology（HIRPO201505） ..................................................................................... 89

Subject: Research on Hadoop & MPP DB using ServerSAN to accelerate performance and

space efficiency in virtualized environment ............................................................................ 89

Subject: Research on Performance Improvement for Open vSwitch Forwarding ................... 93

Subject: Efficient equation rating engine ................................................................................ 97

Subject: Lane level local path planning ................................................................................. 100

Subject: Scalability and Parallel Features of Algorithms and Its Applications in Matrix

Computation ......................................................................................................................... 103

Subject:Researches on Deep Learning based Chinese Image Text Recognition .................... 107

Subject: Analysis and design of programming models of neuromorphic systems with

4 / 206

applications on spiking deep neural nets .............................................................................. 111

BigData Technology（HIRPO201506） ........................................................................................ 115

Subject: Redis Cluster Reliability Improvement .................................................................... 115

Subject: Hybird Big Data Cloud Services Technology Research ............................................. 118

Subject: pan-social data analysis ........................................................................................... 121

Media Technology（HIRPO201507） ........................................................................................... 131

Subject: Development of Internet transmission protocol with high performance in wired and

3G/LTE network ..................................................................................................................... 131

Subject: Research on screen content codec’s performance and coding speed .................... 137

Subject ： Digital Photography, Image Enhancement, Saliency Detection, Composition

Optimization .......................................................................................................................... 139

Subject: Research on AVC-HEVC Transcoder Based on AVC Decoding Information .............. 142

Security Technology（HIRPO201508） ........................................................................................ 144

Subject: Searchable Encryption and Cryptographic Access Control...................................... 144

Subject: Lightweight cryptography performance on resource-constrained devices ............. 150

Subject: Docker Security Research ........................................................................................ 153

Subject: Hardware Device authentication with hardware fingerprint in M2M of IoT .......... 156

Subject: Key Establishment and Generation with Wireless Channel CSI in M2M of IoT ....... 161

Engineering Technology（HIRPO201509） .................................................................................. 166

Subject: New Electrolytes and Electrodes’ Surface Modification in LMB ............................. 166

Subject: Research on energy information physical fusion system ........................................ 168

Subject: Sensor based Innovative interactive application design for Smart phone .............. 171

Subject: Exploring Metamaterial to improve antenna array performance ........................... 173

Subject: Low PIM and High Power Compact wideband coupler using suspended transmission

line ......................................................................................................................................... 174

Subject: Low PIM and High Power Compact combiner using suspended transmission line . 175

Subject: Compact RF filter under moderate power .............................................................. 177

Subject: Novel methods for crosstalk cancellation in high-speed parallel bus ..................... 178

Subject: Novel static and low frequency shielding ................................................................ 179

Subject: Research in Current-Carrying Capacity and Reliability for PCB Traces .................... 180

List of Abbreviations .............................................................................................................. 180

Subject: Research in improving the anti-saturation ability of the active EMI filter .............. 182

Subject: Ultra-high frequency Thin Film power Inductor(TFI) Testing and Modeling

Techniques ............................................................................................................................ 184

Subject: Design,manufacture and performance study technologies for 20MHz+ high

frequency power supplies and low profile magnetic component ........................................ 187

Subject: High transparent flexible OLED realization and innovation application .................. 189

Subject: The pressure sensing TP .......................................................................................... 190

Subject: N in one LGP ............................................................................................................ 191

List of Abbreviations .............................................................................................................. 191

5 / 206

Subject: New type IR Camera ................................................................................................ 193

Subject: Research on the silicon-based SiGe device integration ........................................... 195

Subject: High output power hign operation temperature laser source ................................ 197

Subject: Analog Laser for ROF with 6~10GHz bandwidth at 85oC......................................... 198

Subject: Effect of the metallization on the quality factor of dielectric resonator ................. 199

Subject: Research on Micro displacement during Glue/Solder solidification process .......... 201

Subject: A kind of component which has positive current/temperature coefficient

characteristic ......................................................................................................................... 203

Subject: Fire disaster protection technology for short circuit of lithium battery energy

storage system ...................................................................................................................... 205

6 / 206

Wireless Communication Technology

（HIRPO201501）

Subject No : HIRPO20150101

Subject: Visible light communication

List of Abbreviations

IEEE Institute of Electrical and Electronics Engineers

LED light emitting diode

VLC Visible Light Communication

Background

Visible light communication (VLC) refers to short-range optical wireless

communication using the visible light spectrum from 380 to 780 nm. There has

been increasing interest in visible light optical communication due to the

widespread deployment of light emitting diode (LED)s for energy efficiency and

recent advancements in LED technology. The bandwidth available for VLC is

plenty compare to traditional wireless communication systems. Besides, the

bandwidth is un-licensed. With high bandwidth, VLC has the potential to

realize multi-gigabit-per second data rates over short distances.

Enabled by recent advances in LED technology, IEEE has released 802.15.7

to supports high-data-rate visible light communication up to 96 Mb/s by fast

modulation of optical light sources.

Scope

The research mainly focuses on VLC prototype developments. As a basic

requirement, a VLC prototype that supports PHY of IEEE 802.15.7 is expected

to be developed.

7 / 206

On top of that, new techniques for VLC communication are explored. This may

include using MIMO for VLC communications. The new techniques are

expected to improve the system efficiency or reliability of VLC systems. They

are also expected to be verified using the prototype developed.

Expected Outcome and Deliverables

The expected outcome and deliverables include

VLC prototype that supports

IEEE 802.15.7

Advanced VLC technologies designed during the project

Technical report, include detailed description of the prototype, advanced VLC

technologies designed during the project.

Codes for the simulator and hardware language for prototype development.

Acceptance Criteria

The prototype developed during the project will be evaluated by field testing.

The technical reports and codes will be reviewed by the reviewing committee

of Huawei.

Phased Project Plan

Phase Duration Tasks Outputs

1 August 1st ,

2015 –

February

1st, 2016

Analyze IEEE 802.15.7

specifications. Design

the transmitter and

receiver algorithms.

Perform link level

simulations. Purchase

the hardware and

Link level simulator.

Technical reports for

algorithm description.

8 / 206

software needed for

prototype development.

2 February

1st, 2016 –

August 1st,

2016

Develop prototype,

realize 802.15.7 PHY I

in the prototype.

Prototype; codes;

specifications for the

prototype.

3 August 1st,

2016 –

February

1st, 2017

Design advanced VLC

technologies, e.g.

MIMO, or application of

VLC in intelligent traffic

system. Implement the

designed algorithm on

the platform.

Prototype; codes;

technical report.

4 February

1st, 2017 –

August 1st,

2017

Design advanced VLC

technologies, e.g.

MIMO. Implement the

designed algorithm on

the platform. Conclude

the project.

Prototype; codes; final

technical report.

Phased Project Plan

two year

9 / 206


Subject: Design Massive MIMO for Energy Efficiency


MIMO: Multiple-input Multiple- output

EE: Energy Efficiency

Background

Energy efficiency is one of critical goals of future network. Massive MIMO

regarded as key approach, can achieve higher EE than single antenna system,

and obviously improve the network coverage and capacity, and is expected to

improve EE as well.

Whether Massive MIMO is energy efficient still remains open. From an energy

efficiency perspective, it utilizes the large antenna array to form narrow beams,

and accurately put the directional power to users. Therefore, the overall

transmission power can be significantly reduced, However, Massive MIMO get

the above gain at the cost of extra radio frequency (RF) links and complicated

signal processing, both of which considerably increase power consumption of

the system.

So, it is a valuable research direction to analyze the EE of Massive MIMO and

to design Massive MIMO to achieve maximal EE.

Scope

Research on energy efficiency model of Massive MIMO system:

investigate the parameters and factors which affect the energy efficiency of the

system, build the energy efficiency module of the system;

Research on energy efficiency solution design: based on the energy

efficiency model, design algorithms how to choose the number of the base

station antennas, the number of users and combine power allocation with

10 / 206

respect to each antenna to achieve the maximum EE in all kinds of typical

scenarios (eg. density urban, urban)


Technical reports of energy efficiency model and analysis for Massive MIMO;

Technical reports of energy efficiency solution design, including theoretical

optimal system parameters analysis, candidate schemes of optimal resource

allocation and adaptive resource adjustment scaling with traffic ,the

performance simulation of the scheme;

Massive MIMO simulation platform with source codes and description;

1~2 Invention/patents;

Phased Project Plan

Phase1 (~3 months): survey the state of the art of massive MIMO in energy

efficiency field, analyze and build the energy efficiency model and provide the

related technical report.

Phase2 (~5 months): Research on system parameters design based on

energy efficiency model to achieve the optimal EE in kinds of typical scenarios

and provide the related technical report.

Phase3 (~4 months): Research on schemes of adaptive resource adjustment

scaling with traffic and provide related algorithms, simulation results and

patents.

11 / 206


Subject: Reconstruction or approximation to FDD

Downlink Channel Information


Background

In FDD system, due to frequency gap, downlink channel responses are

commonly deferent with uplink channel, i.e., reciprocity is not available for

FDD.

Otherwise, it's often considered that in most cases, uplink signals propagate in

almost the same way as downlink signals. The angle of arrival and departure

could be an example.

Some papers suggest that if the main components of one direction can be

drawn correctly, the other can be reconstructed with tolerable errors.

If we get more channel information of downlink, especially from the uplink

signals such as SRS transmitted or CSI fed back by UEs, we can try to

approach more gain of MIMO in LTE FDD system, despite the loss caused by

limited size of precoding book.

Scope

The modeling of LTE FDD downlink and uplink channel in typical scenarios;

including supporting document and simulation platform;

The theoretical analysis of reconstruction of downlink channel based on limited

uplink channel information, including the bound analysis;

The solution and algorithms to reconstruct or approximate the main part of

downlink channel or eigen information in legacy LTE FDD system, based on

uplink signals or feed back; which could be verified on LTE product platform

12 / 206


Publications in peer-reviewed Journals or top ranked conferences, and/or

invention/patents.

Technical reports of algorithm design, including theoretical capacity limitation,

investigation of candidate schemes, the design and performance analysis of

the new schemes.

Simulation platform with the source code and simulation results.

Acceptance Criteria

The reconstructed or approximated LTE FDD downlink channel information

can be used for single user beamforming and multi-user beamforming with

slight capacity loss (less than 10-15%) compared with the idea case, i.e., the

eNB has the full downlink channel information;

No needs to have LTE FDD UE feed back more besides legacy CSI;

No needs to insert uplink pilot at downlink frequency, or other tech to modify

the LTE standards.

Phased Project Plan

3 Phases.

13 / 206


Subject: High reliable wireless network technology for

buildings


Internet of Things (IoT)

Background

China will very soon become the No.1 owner of high rise and super-tall

buildings all over the world. When disaster happens, it is great challenge for

people to escape without the evacuation guidance. In order to improve the

evacuation guidance, a key challenge is to establish a reconfigurable

information network, which can maintain the connectivity under node failure

and link failure. In existing buildings, the communication network is mainly a

hybrid, which consists of wired network that spans to each floor, and wireless

network that serves the users in several zones in the same floor. In high-rise

buildings, the wired network becomes vulnerable because the communication

path across the levels may be cut off by disaster or infrastructure failure. On

the contrary, wireless network has the potential to maintain integrity under both

node failure and link failure.

The focus of this project is to design a reconfigurable Internet of Things for

buildings, which connect the sensors, actuators, and human-building interface

through mainly wireless network. This reconfigurable IoT shall maintain

integrity under extreme conditions and therefore enable smart evacuation

guidance.

14 / 206

Scope

Fig. 1 Reconfigurable Internet of Things for Smart Evacuation Guidance in

Buildings

The project includes three parts:

i. Use wireless network to connect popular sensors, actuators, and

human-building interface in ordinary buildings.

ii Design reconfigurable strategies under scenarios of node failure and link

failure.

iii A simulation/experimental demo within a given area in a building.


i. A research report

ii. A simulation/experimental demo

iii 1 patents

15 / 206

Acceptance Criteria


ii. A simulation/experimental demo

Phased Project Plan

i. Network: 4 months

ii. Reconfigurable strategy design and test: 6 months

iii Research report: 2 months

16 / 206


Subject: Capacity Analysis of SCMA Systems


Background

The advent of smart phones and tablets over the past several years has

resulted in an explosive growth of data traffic over the cellular network not

seen in previous generations. With the proliferation of more smart terminals

communicating with servers and each other via broadband wireless networks,

new applications emerge to take advantage of wireless connectivity. Fifth

generation (5G) wireless networks are expected to support very diverse

services from very low latency to very high delay tolerant, and from very small

to very large packets in different applications. An important 5G requirement is

that it should support massive connectivity with a large number of devices

such as smart-phones, tablets and machines. As a result, multiple access

schemes play a critical role in providing the increasing demand in services for

future terminals and applications. The current Long Term Evolution (LTE)

system is not able to efficiently support massive connectivity, especially in the

uplink. Among the issues that arise due to massive connectivity is the cost of

signaling overhead and latency.

Sparse code multiple access (SCMA) is a new multiple access scheme. SCMA

is a codebook-based non-orthogonal access mode allowing overloading of the

system with a large number of SCMA layers to enable massive connectivity.

Sparsity of SCMA codewords makes the near-optimal detection of over-laid

SCMA layers practically feasible. It’s reported that SCMA can provide around 3

times gain over OFDMA in terms of supported active users. Therefore, SCMA

is a promising technology for 5G wireless networks for data transmission with

low signaling overhead, low delay, and support of massive connectivity.

17 / 206

Scope

This research focuses on the capacity analysis of SCMA system: given a

specific time-frequency resource block and minimum transmission rate for

each user, what is the maximum number of users which can simultaneous

access an SCMA system?

This problem is in fact an information theoretical problem of multiple access

channel (MAC) when considering infinite code length, ideal capacity-achieving

error-correcting coding, static channel state.

It turns to be much more complex when considering some practical setups in

SCMA system. For example:

The error-correcting capability is limited due to the code length;

The channel state is no longer static but time-varying, i.e., fading channels;

The maximum achievable transmission rate is limited by number of modulation

constellation points.

A capacity or performance bound under the above practical considerations will

be very helpful for optimizing the SCMA system.


Papers and research reports, and related programming source codes.

Acceptance Criteria

At least 2 technical reports and 1 related publication.

Phased Project Plan

Phase 1: Derive capacity of SCMA system under ideal settings;

Phase 2: Derive capacity of SCMA system under fading channels;

Phase 3: Derive capacity of SCMA system with finite code length;

18 / 206

Phase 4: Derive capacity of SCMA system with finite number of constellation

points.

19 / 206


Subject: Wireless signal based gesture recognition

technology


Background

Along with the development of wifi and other radio technology, the WiFi access

point are ubiquitous in private home and public places. Because the human

body is the good conductor of electricity, the human has strong reflex action to

the radio wave. 802.11n-based and 802.11ac technology the WiFi commercial

chip may precision measuring human body's perturbation to the wireless signal.

Thus, the WiFi radio equipment may detect and recognize the human body

movement and gestures。

Scope

The scope of this project includes but not limited to:

1. Detailed analysis on the working principle and implementation mechanism

of Wireless signal based gesture recognition technology.

2. Implementing a prototype for Wireless signal based gesture recognition

technology.


1. One patents;

2. One paper;

3. One technical report with respect to the design and implementation;

4. Prototype for the proposed methods.

20 / 206

Acceptance Criteria

Fail: No patents/papers/prototype are delivered.

Pass:

1 patents pass Huawei’s review AND

1 paper is accepted at a IEEE flagship conference/transaction AND

1 detailed technical report AND corresponding prototype

Excellent:

More than 1 patents are delivered, OR

More than 1 paper is accepted by IEEE flagship conferences/transactions.

Phased Project Plan

Phase 1: T+4, deliver a patent and a survey on Wireless signal based gesture

recognition technology;

Phase 2: T+8, deliver one technical report;

Phase 3: T+12, deliver one paper, and prototype.

21 / 206

Future Networks（HIRPO201502）


Subject: Research on Internet of Things (IoT) carrier

network architecture and technology


Background

The Internet of Things (IoT) refers to the interconnection of uniquely

identifiable embedded computing-like devices within the existing Internet

infrastructure, which offers advanced connectivity of devices, systems, and

services between machine-to-machine and human-to-machine, and covers a

variety of protocols, domains, and applications. Recent years, as promoted by

sensor, big data and mobile computing technologies, the global IoT market is

growing rapidly, reaching 350 billion USD by 2015 as predicted, which

promotes the overall profit of global enterprises by 21%.

IoT applications have their unique characteristics compared with traditional

Interne applications: it produces massive heterogeneous data by a wide

variety of devices such as heart monitoring implants, biochip transponders on

farm animals, automobiles with built-in sensors, or field operation devices that

assist fire-fighters in search and rescue; it is applicable in nearly all fields,

through various transportation and communication mechanisms, combined

with the power of cloud computing and the intelligence of big data analysis.

IoT brings lots of opportunities, and in the meantime, it brings many challenges.

There is a variety of IoT applications which have different traffic patterns and

requirements on network. How those traffic models influencing on current

carrier network (includes access networks, core networks, data center

22 / 206

networks) architecture? What is the optimal network architectures according to

heterogeneous consumption patterns?

Security represents a critical component for enabling the widespread adoption

of IoT technologies and applications. Both the range and the depth of the

influencing, that the IoT industry brings more controls and intelligent, on our

everyday life. There will make very serious threaten when IoT application be

attacked than ever before. What are the major security challenges of the IoT

technology? What the security considerations the carrier operators should

have? How to re-construct carrier network to adapt the IoT applications

security?

Nowadays, operators’ business model is still limited in charge with connection

or traffic, how should the operator to do to make innovations in new business

model and to increase share of revenue in IoT ecosystem?

The answers to those fundamental questions will have profound and valuable

impact on the current operators business strategy as well as the evolution of

the future carrier network.

Scope

The scope of this project covers various types of IoT applications and

scenarios, e.g. Vehicle-to-Vehicle communication, live video monitoring, smart

building, smart home, industrial 4.0, and etc. And how all above IoT industries

impact on the carriers’ networks. The content of the project includes, but is not

limited to, the following parts:

Analysis of IoT applications’ traffic characteristics, modeling those traffic

consumption according to various scenario, and research how those traffic

make influence on carrier network architecture.

23 / 206

Analysis of IoT application requirements on carrier network, and how the

network could do to adapt to those requirements by providing new services,

technologies, and other methods.

Analysis if the current network security services could fulfill the IoT security &

privacy requirement, and find out any chances which is not limited to new

technical, new services and solutions for carrier operator.

Analysis how will IoT applications involve in SDN&NFV network innovation.

Analysis how operators do to gain more revenue, and make more impact in IoT

ecosystem, by making innovation on business models.


The deliveries of the project include, but are not limited to the following:


invention/ patents on IoT impact on carrier network, network related

technology innovation, IoT network security, or SDN&NFV correlation with IoT

network

Research reports on IoT applications traffic patterns, scenarios, new carrier

network architecture, network security, etc.

Prototype on IoT based future network architecture, with source codes and

corresponding design and implementation documents.

Acceptance Criteria

One or two papers published in peer-reviewed Journals or top ranked

conferences, and/or invention/ patents on IoT impact on carrier network,

network related technology innovation, IoT network security, or SDN&NFV

correlation with IoT network.

At least one open speech or demonstration in an IoT top level meeting or

associated workshop.

24 / 206

The research report or thesis about IoT network’ security, IoT traffic’ analysis,

IoT industry’s impact on every kind of network areas(including access network,

transport network, core network, data center network), new business model,

and so on, should be accepted by Huawei’s Review Group.

Phased Project Plan

Stag

e

Date Work description Output Evaluation Criteria

1 T*~T+2M

onths

Specify milestones.

Thesis proposal cover

the whole research

scope.

Routine technical &

work progress meeting.

1, An determined

work plan about

what should we do

in this project and

how to guaranty the

successful of the

collaboration

2, Thesis proposal

3, Research report

on one or more

items described in

section 5

The documents can

be accepted by

Huawei’s Review

Group.

2 T+3

Months ~

T+7

Months

Continuing the research

work.

Academic paper’ writing.

Prototype design and

coding.

1, Research report

on more items

described in section

5

2, Complete at least

one academic paper

The design

documents can be

accepted by

Huawei’s Review

Group.

25 / 206

Routine technical &


3, Prototype design

document & source

code(partial )

3 T+8

Months ~

T+12

Months

Complete the research

work.

Academic paper is

accepted by the

Journals or top ranked

conferences.

Implement prototype for

demonstration and

verifying.

Routine technical &


1, Research report

on all items

described in section

5

2, Complete all

papers

3, Complete

prototype

1, Finish the

prototype

implementation,

complete the

prototype’s coding,

testing, verifying, and

related report

2, Hold an associated

workshop or attend a

IoT related summit,

on which make an

open speech or

demonstration.

3, The paper

published in

peer-reviewed

Journals or top

ranked conferences

26 / 206


Subject: Bearer Network for 5G

To investigate breakthrough architecture and the enabling technologies to

satisfy the 5G core KPI including the fronthaul / backhaul and the core network

excluding air interface technologies.


5G：5th generation network

EPC：Evolved Packet Core

MM：Mobility Management，steering the traffic to the mobile node wherever it

moves.

Background

5G is the next generation wireless network and is the biggest thing in the

communication industry. The core KPIs of 5G (1ms latency, 1G-10Gbps

bandwidth, 1 million connections per km square) would change the network

architecture as well as new air interface technologies. Distributed EPC is

required to shorten the latency while avoiding traffic detouring caused by

distributed EPC. The current wireless core network inherits the architecture

from voice time, i.e., TDM-Like mobility management. It setup/update tunnels

between core entities and base stations when mobile nodes move. It has

potential architecture limitation in terms of scalability and latency performance.

Breakthrough architecture and the enabling technologies are worth

investigating to satisfy the challenging 5G KPI.

Scope

Two categories of 5G network need to be taken into consideration:

backhaul/fronthaul which connects the base stations to core network and the

27 / 206

core network itself which does the mobility management, i.e., steering the

traffic to the moving nodes.

Acceptance Criteria

All the deliverables should pass the expert review. To meet this goal, the study

should include both quantitatively and qualitative proof for the required items

including traffic model studying, the architecture proposal and enabling

technologies solution.

Expected Outcome and Deliverable

5G network challenges report

5G xhaul network proposal , 5G core network proposal

5G network demo

Phased Project Plan

Phase I: Problem Identification, Sep 2015 - Dec 2015

Understand the 5G requirement, study the traffic model of fronthaul/backhaul,

identify the potential architecture limitation of current wireless core network in

terms of mobility management.

Phase II: Architecture Proposal, Jan 2016 – Mar 2016

Propose the network architecture and enabling technologies including

fronthaul and backhaul network and the core network with breakthrough

mobility management.

Phase III: Enabling technologies design, prototype development and

verification, Apr 2016 – Dec 2016

Develop the prototype and enabling technologies and verifying the prototype

by either dry run or simulation.

28 / 206


Subject: Research on in network computing


Networking computing

Background

Collective operations, such as all reduce, are widely treated as the critical

limiting factors in achieving high performance in massively parallel applications.

Conventional host-base implementations, which introduce a large amount of

point-to-point communications, are less efficient in large scale systems. To

address this issue, we propose a design of switch chip to accelerate collective

operations.

Compare with the NIC-based allreduce, the switch based allreduce

furthermore improves the performance of the allreduce operation. Both the

processor and the NIC participate in the operation only when they send the

initial data to and receive the result from edge switches, which greatly reduces

the amount of communication. Since each node only communicate with

switches rather than other nodes, the communication path is shorter, which

expects lower latency.

Scope


1. Detailed analysis and implementation mechanism of network computing.

2. list offload points for networks computing, include barrier

3. realize and simulate, give the simulation/results

4. performance optimize of network computing.

29 / 206


1. Two patents;

3. One technical report with respect to the design and implementation of

networks computing;

4. Simulation/test results for the proposed methods.

Acceptance Criteria

Fail: No patents/papers are delivered.

Pass:


1 detailed technical report AND corresponding simulation/test results.

Excellent:


Performance exceeding expectation

Phased Project Plan

Phase 1: T+4, deliver a patent and a survey on network computing

Phase 2: T+8, deliver a patent; realize and simulate

Phase 3: T+12, deliver test results, one technical report and simulation/test

results.

30 / 206


Subject: Research on DCE NIC offloading


MPI: message passing interface

DCE: Data Center Ethernet

Background

DCE NIC scenarios include: HPC, BigData, Cloud. In HPC field, for the most

widely used MPI program, bring MPI offload technology to NIC; in the other

two scenarios: BigData and Cloud, whether or not the same specific design,

through the NIC offloading technologies to improve the performance of

applications ?

Scope


1. detailed analysis and implementation mechanism of DCE NIC offloading.

2. describe benefit of DCE NIC offloading

3. design DCE NIC offload mechanism

4. realize and simulate, give the simulation/results

5. performance optimize of DCE NIC offloading.


1. Two patents;

3. One technical report with respect to the design and implementation of DCE

NIC offloading;


31 / 206

Acceptance Criteria


Pass:



Excellent:


Performance exceeding expectation

Phased Project Plan

Phase 1: T+4, deliver a patent and a survey on DCE NIC offloading

Phase 2: T+8, deliver a patent; realize and simulate

Phase 3: T+12, deliver test results, one technical report and simulation/test

results.

32 / 206

Optical Technology（HIRPO201503）


Subject: RF Crosstalk Analysis in CFP2-DCO module


ACO analog coherent optics

CFP C form-factor pluggable

DCO digital coherent optics

DSP digital signal processing

GND ground

PCB printed circuit board

PM-QPSK polarization-multiplexed quadrature phase-shift

keying

PIC photonic integrated circuits

RF radio frequency

TIA trans-impedance amplifier

Background

Driven by the exponentially increasing customer demands for broadband

services, high speed coherent technology has emerged as a promising

candidate to provide huge transmission capacity. Nowadays, 100-Gbps

PM-QPSK has been widely deployed all over the world. However, the previous

optical module has large size and high power dissipation. To obtain compact

transceiver and lower power consumption, the downsizing of optical module

based on photonic integrated circuits (PICs) is indispensable. CFP-2 has been

proposed as a standard size (107.5 x 41.5 x 12.4 mm3) for next generation

33 / 206

optical module, where integrated optical devices and electrical devices are

required.

Along with downsizing of the device footprint and module size, some serious

problems need to be overcome. Among these problems， RF crosstalk issue in

high-density integrated module needs to be addressed, where the four

different signals with 34-Gbaud are amplified by driver and the received four

signals are amplified by TIA. Since the electrical signal has higher frequency

and larger amplitude, signals in one chip and from different chips can result in

significant crosstalk in the small-sized module.

Therefore, it is critical to analysis the RF crosstalk in integrated optical module,

which can reduce the research risk. Our target is to provide optimal solution to

minimize RF crosstalk in next generation small-sized optical module.

Scope

CFP2-ACO module has been produced by Oclaro and Finisar. For next step

CFP2-DCO module, smaller optical and electrical devices are required since

the optical DSP chip will occupy a lot of space in the module.

*The requirements can be modified after final discussion.

This project needs to solve the following problems:

1 In a small-sized module or chip, the four high-frequency electrical signals

with large amplitudes at Tx side can significantly effects the four

high-frequency electrical signals with small amplitudes at Rx side due to the

RF crosstalk. A schematic needs to be provided to reduce the RF crosstalk by

using isolation, optimal layout, and other methods.

2 For the chip bonding pad and power supply design of PCB, the positions and

numbers of bonding pads need to be optimized. The optimal solution for

practical application scenarios should be solved by using the models.

34 / 206


1、Two research reports 2、Simulation models 3、Two patents

Acceptance Criteria

Outcome and deliverables will be reviewed, evaluated and accepted by

committee of Huawei for this project including 3~4 experts in related realms.

Phased Project Plan

Phase Duration Tasks Outputs

1 T0(contract

signed)-T0+4months

A schematic needs

to be provided to

reduce the RF

crosstalk by using

isolation, optimal

layout, and other

methods

Simulation models

2 T0+5months –

T0+8months

For the chip bonding

pad and power

supply design of

PCB, the positions

and numbers of

bonding pads need

to be optimized. The

optimal solution for

practical application

scenarios should be

solved by using the

models.

Simulation models、

research report

35 / 206

3 T0+9months –

T0+12months

Simulation models

and solutions to

reduce RF crosstalk

are optimized and

support Huawei to

apply patents.

Research report 、

patents materials

36 / 206


Subject: All optical metro network and key technology

collaboration


Background

Today, Web 2.0 users not only consume content, they also generate content.

Anyone can publish content on the Internet and user generated content (UGC)

has become mainstream information. In 2009, 75% of the content published on

the Internet was generated by users. The type of content users generate is

changing as well. Not only do more and more people own video or SLR

cameras that can produce high definition videos, it is increasingly easy to

publish these videos on the Internet. Users are generating an enormous

amount of digital content and the amount of information available is literally

exploding. In the coming decade, experts estimate that information available

online will increase 270 times. Processing all of this data with require massive

computing and storage capabilities. In addition, estimates project as many as 5

billion mobile broadband users, with each user consuming 50% more traffic

year to year. Total network traffic will increase 500 times over the course of ten

years. This information explosion and the digital floods result present network

carriers with unprecedented challenges.

37 / 206

Figure.1 Fast growth of service traffic caused by globalization of information

Currently, the evolution on video era is around 4K and 8K. NHK Japan plans to

achieve 8K live in 2020. Europe Pay-TV operators, including telecom

operators and MSO in positive action, in order to achieve commercial 4K final

sprint during the 2016 European Cup. To meet the business experience to

meet the high 4K, 4K bearer network needs to make critical requirements.

For telecom operators, metro and backbone bandwidth to upgrade to 100G +,

the current network levels, and multiple network hops cause large network

delay for 50 ~ 100ms, in addition to the delay, today's carrier networks throw

package rate 10-4; 4K MAN made low latency, high bandwidth, very simplified

requirements.

Standard Quasi 4K Basic 4K IMAX 4K

Resolution 3840*2160 3840*2160 3840*2160

Frame Rate 25P 50/60P 100/120P

Sampling Bits 8 10 12

Compression HEVC main profile HEVC main 10 HEVC Range Extension

Average bit rate 15~20Mbps 30~50Mbps 50~100Mbps

Bandwidth Requirement

(1.5 x Average Bit Rate) 20~30Mbps 50~75Mbps 100~150Mbps

38 / 206

The advent of 5G technologies and ICT networks signify the coming next wave

of a globally connected Digital Society:

Right now, all over the world, mobile access to the internet is becoming wholly

fundamental to doing business in all industries. Flexible working practices

facilitated by mobile networks and devices are already essential, and are

allowing enterprises to conduct operations across boundaries that previously

inhibited growth.

Current Network Requirement of Basic 4K

Channel

Change Time

1.5s 0.6s

Throughput 23M 45M

RTT 50~100ms 35ms (OTT)

PLR 10-4

5*10-5

(OTT)

5.6*10-7

(BTV)

Access

Bandwidth

<50M Basic: 45M

Burst: 60M

Traditional Network

Core

Metro Ring

UPE UPE

AGG AGG

SR BRAS PE

UPE

39 / 206

Growing mobile access to the internet, cloud-based services and Big Data

analytics is allowing anyone, anywhere to leverage “Big Wisdom” – a whole

new kind of globally connected and shared knowledge base.

The continuing rise in the relevance of social media as an important part of

how we interact with the internet is also opening up new kinds of intelligent

analytics ready to be harnessed for tangible business and everyday life

benefits.

Transformation and convergence of ICT network infrastructure is driving

business innovation and growth. Not only is ICT an increasingly effective tool

for enhancing efficiency, but it is now a vital driver of economic and societal

growth.

The development of 5G technologies is a cornerstone for

realizing breakthroughs in the transformation of ICT network infrastructure.

Ultra-broadband and intelligent-pipe network features that achieve

near-instantaneous, “zero distance” connectivity between people and

connected machines – no matter where they are – are just the first step.

An increasingly diverse and wide range of mobile services will have differing

performance requirements:

Latency from one millisecond to a few seconds

Always-on users per cell from a few hundred to several millions

Duty cycles from mere milliseconds to entire days

Signaling loads from less than 1% to almost 100%

The “5G HyperService Cube” below gives a multi-dimensional overview in

terms of throughput, latency and number of connections required for the many

types of services 5G networks will need to run:

40 / 206

5G networks faces significant design challenges to simultaneously meet all of

the above service requirements. They must be built to meet a number of

individual user and enterprise needs:

Immersive experience: at least 1 Gb/s or more data rates to support ultra-high

definition video and virtual reality applications

Fiber-like user experience: 10 Gb/s data rates to support mobile cloud service

Zero latency and response times: less than one millisecond latency to support

real time mobile control and vehicle-to-vehicle applications and

communications

Zero-second switching: max 10 millisecond switching time between different

radio access technologies to ensure a consistently seamless delivery of

services

Massive capacity and always-on: current mobile network systems already

support 5 billion users, this will need to expand to also support several billions

of applications and hundreds of billions of machines

Energy consumption: energy-per-bit usage should be reduced by a factor of

1,000 to improve upon connected device battery life

41 / 206

In response to these new requirements, the traditional transport network in

exchange capacity, energy conservation, equipment and other aspects of the

spatial volume of the difficulties faced. High capacity single-system power

density is too large, difficult to deploy, China Mobile to give the red this year

consumption more than 10kW / rack is not allowed to deploy. Large-capacity

equipment Evolution encountered two major bottlenecks, high-speed links

bottleneck, 30G NRZ (1T) has PCB backplane limit is reached, 56G (2T) PAM4

difficulties, back plate further room for improvement has been very limited and

can not use cable technology to enhance link performance, single-box optical

backplane and chip optical fiber technology is not mature, equipment power

bottlenecks, single power is growing, more and more difficult to deploy

equipment; the case of doubling capacity, generation chip technology

advances, chip power consumption will continue to increase 20% ~ 40%,

16nm in the future, more and more investment income is lower than the chip,

the new chip development rhythm slowed, reducing power consumption and

more difficult.

Figure.2 Power evolution of single Rack with 100T switching capacity

Aligning with this trend, all-optical network is a potential promising solution,

which facilitate sub-wavelength and greatly reduce electrical-layer processing

at intermediate nodes. As shown in 0, unlike present network architecture in

which the traffic is aggregated and switched at Layers 0, 1, and 3, AON

support sub-wavelength provisioning and switching directly at the optical layer.

These networks have the potential to reduce and eventually eliminate the

42 / 206

costly O/E/O conversions at each switching nodes, leading to great reductions

in deployment costs and power consumption.

Figure.2 Current network switching architecture vs. all-optical switching

architecture

Scope

The metro optical network studied in this project is divided into metro

aggregation and metro

core, the physical distance covered 100km ~ 600km. The whole network

includes 8 core node, which should have 100Tbps switching capacity and

energy consumption should be less than 10mW/Gbps. When the signal enters

the network from the network edge, it should be always in the optical domain

whenever it’s transmitted or switched, to guarantee the E2E latency is less

than 10ms.

The main purpose of this project is to build an all optical network architecture,

including key building blocks which has functions like optical signal generating,

43 / 206

transmitting, switching, processing, and protocols to guarantee high capacity,

low power consumption and low latency.

The research can include, but not limited to:

The overall architecture of all optical metro network, which can satisfy the

traffic requirement and realize fast light path setup (setup time less than

10ms).

The key building block models and realization solutions

Optical signal generator. Covering C band, the optical channel bandwidth in

metro core >400Gbps, in metro agg >100Gbps, and keeping the same SE with

traditional network. The sub-wavelength in optical channel can be 10Gbps

High capacity optical switch/router. Supporting max 12 dimensions,

sub-wavelength switching granularity, and capacity up to 100Tbps, power

consumption less than 10mW/Gbps.

Low cost wavelength converter, either in optical way or electronic way

Optical performance monitor and fast light path setup method.


The expected outcome and deliverables are listed in the table:

Deliverables

1 The overall architecture design report of all optical metro

network

2 The key building blocks design report part1

3 The key building blocks design report part2

44 / 206

Acceptance Criteria

In this project, the new all optical network architecture and solution should

keep the same SE, network throughput, and network effcienty (cost per Gbit)

as traditional network. The core node capacity should be at least 100Tbps, and

the power consumption should be <10mW / Gbps. The E2E delay should be

less than 10ms.

Phased Project Plan

Stage 1 2 3

Date

(from-until)

T*~T+3 Months T+4 ~T+8

Months

T+9 ~T+12

Months

Work

Description

To explore

possible new

optical network

architecture that

can satisfy the

requirement

described in

section 5 Scope

and section 7 .

To explore the

key building

blocks

realization of

corresponding

new optical

network

To explore the

key building

blocks realization

of corresponding

new optical

network

Output The overall

architecture

design report of

all optical metro

network

The key

building blocks

design report

part1

The key building

blocks design

report part2

Evaluation

Criteria

The document

can be accepted

by Huawei’s

The design

document can

be accepted by

The design

document can be

accepted by

45 / 206

Review Group. Huawei’s

Review Group

Huawei’s Review

Group

46 / 206

Storage Technology（HIRPO201504）


Subject: Research on New programming model based

on NVM


NVM: Non Volatile Memory

Background

Trend

In recent years, NVM (such as RRAM、PCM、MRAM、STT RAM, etc)have

made great process, their persistence, high capacity, high density, low power

consumption, high performance and long wear life cycle cause academia and

industry attention, they bring us hope for enhancing the storage system

capabilities in cloud computing and large data background.

Challenge

NVM is a relatively new medium, we know quite little about its characteristics.

Researchers have made some preliminary research about NVM. For example,

using NVM as a general block device, building file systems on NVM，

enhancing storage array performance with NVM, etc. However，all the trial do

not take full advantage of NVM，and there are many usage modes for NVM，

how to better use these modes is lack of research.

Value and Objectives

NVM is developing very fast，it will be quite helpful for enhancing storage

system capabilities in the future, we can quickly get accumulation of

experience and core technologies in NVM field by technical cooperation.

47 / 206

We hope to build a New programming model based on NVM，this New

programming model can provide ACID guarantee , solve the problem of data

consistency, guarantee NVM’s performance\ longevity\cost to achieve optimal

cost-effective, and can help the application to build, organize and access a

number of data structures.

Scope

1. Deeply analyze and verify NVM characteristics

2. Design a New programming model based on NVM to achieve optimal

cost-effective for NVM’s performance, longevity, cost and data reliability

3. Output prototype code and verify it.

4. Output 2 patents and 2 papers


1 Design documents

a) Architecture design document

b) New programming model based on NVM design document

2 New programming model based on NVM Codes

3 NVM simulation platform

4 Test and analysis report

5 Patents

Acceptance Criteria

1. Design document can be approved by review committee:

a) A feasible and detail design description how to build a New programming

model based on NVM

48 / 206

b) A feasible and detail design description how to provide ACID guarantee ,

solve the problem of data consistency, optimize each level Cache including

DRAM to reduce clflush/fence/cache miss numbers, guarantee NVM’s

performance\ longevity\cost to achieve optimal cost-effective, and how to

help the application to build, organize and access a number of data

structures

2. Test and analysis result can meet the specification:

a) If IOPS=1M and 4KB IO, longevity: ≥ 5 years

b) If power down or reboot or system crash, no data loss

3. At least 2 patents and all patents can be approved by IPR

Phased Project Plan

T: Time of signing contract

1. Phase 1

1) Time: T + 0 months ~ T + 2 months

2) Working description:

a) Make clear the deliverables and acceptance criteria

b) Complete the requirements analysis

c) Complete the NVM trend/characteristics and use mode analysis document

3) Outcome

a) deliverables and acceptance criteria memo

b) requirements analysis document

c) NVM trend/characteristics and use mode analysis document

2. Phase 2


49 / 206


a) Complete architecture design document

b) Complete New programming model based on NVM design document

3) Outcome

a) Architecture design document

b) New programming model based on NVM design document

c) Test strategy

3. Phase 3



a) Complete prototype code

b) Complete prototype code debug

c) Compete patents idea

3) Outcome

a) Prototype code

b) Patents idea

4. Phase 4



a) Optimize and tune prototype code

b) Complete test and analysis report

c) Compete patents disclosures

50 / 206

3) Outcome

a) Optimized prototype code

b) Updated design document

c) Test and analysis report

d) Patent disclosures

51 / 206


Subject: Storage optimization Research for NoSQL

DBs


NoSQL: Not only SQL

DB: Database

RDBMS: relational database management system

KV: Key Value

Background

NoSQL distributed storage technology is on a rapid rise for Web 2.0 as well as

for a variety of Enterprise applications, not only due to its simplicity and

scalability features by design but also for its distributed nature for the

convergence of data storage and data management. For the above reasons,

NoSQL is capable of supporting real-time applications with a large number of

application users.

In Fig. 1, we show a list of some of the most notable NoSQL vendors in the

world, most of which are headquartered in the US. Those include some of the

document-based NoSQL DBs (MongoDB, Couchbase, etc.), KV-based NoSQL

DBs (Memcached, Redis, etc.) as well as graph-based NoSQL DBs (Neo4j,

etc.).

We illustrate some of the major technical components of NoSQL DB in Fig. 2.

These include cluster management, distributed index and query,

cross-data-center replication, load balancing, etc. Some of the major concerns

in terms of NoSQL DB performance are data integrity and data reliability issues,

compared with some of the leading RDBMS systems.

52 / 206

Fig. 3 shows the top 10 most popular DB engines by March 2015. As we can

see that three of the NoSQL DBs are in the top 10, which are MongoDB,

Cassandra and Redis. The trends also demonstrate that the ranking of all of

those three NoSQL DBs in the top 10 most popular DB engines are still on the

rise.

Fig.1: A Variety of Options for NoSQL Distributed Storage

Fig.2: An overview of NoSQL distributed storage

53 / 206

Fig. 3: An overview of DB ranking in terms of popularity by March 2015

So Huawei is looking for the partners to conduct extensive experiments and

thoroughly evaluate the performance of some of the major NoSQL DBs in

terms of data integrity and data reliability in a variety of scenarios. In what

following, the NoSQL DBs should include MongoDB, Cassandra and Redis.

Scope

The experiments, evaluation and protocol enhancement prototype

implementation should include:

1 The setup of a testbed with at least half a dozen of cluster nodes for the

experiments.

2 The setup of different scenarios, such as different number of nodes,

different workloads, different failure modes.

3 The experiments to examine data loss scenarios in NoSQL DBs and the

concrete likelihood of data loss occurrence for different NoSQL DBs in

different components such as the data replication protocol, data

consistence protocol, the fsync() process, etc.

4 The proposed approach to alleviate or fix the data reliability issues such as

the enhancement of the protocols and the enhancement of fsync(), etc.

54 / 206

5 The implementation of the prototype for the enhancement of the protocols

in NoSQL DBs to alleviate or fix the data consistency and data reliability

issues.

6 The patent ideas for the prototype.


1 A comprehensive experimental and analytical report, describing the

insightful findings regarding data reliability issues in NoSQL DBs in a

variety of scenarios

2 A software prototype and source code to verify the proposed enhancement

protocols.

3 1~2 patent ideas for the prototype.

4 A final report to describe above all.

Acceptance Criteria

1 The experimental results can be reproduced.

2 The report should present the merits/drawbacks, applicable scenarios and

the reasons for data loss in different NoSQL DBs.

3 The software prototype should reduce the data loss ratio by 10%-30%.

4 The patent ideas should be accepted by patent department of Huawei.

Phased Project Plan

Total: 1 year

1 Stage 1：(1 month)

Make a project plan and research approach

2 Stage 2：(6 months)

55 / 206

Setup a testbed and different scenarios for experiments.

Test the different NoSQL DBs and analyze the reasons for data loss in certain

scenarios.

3 Stage 3：(5 months)

Implement a software prototype which can reduce the data loss in the above

scenarios.

Get 1~2 patent ideas for the prototype.

56 / 206


Subject: Research on Distributed Transaction

Management Framework and Prototype


DTM （Distributed Transaction Management）

Background

With the evolution of information technology, more and more systems are

trying to be more scalable and robust by exploiting distributed technologies.

The basic distributed components and framework are key foundations for

distributed systems. Huawei is seeking the cooperation opportunity on the

research and development of distributed transaction management framework

which is used to standardize and ease the development of distributed

transactions.

Scope

requirement scope: description

To implement a full

function of

distributed

transaction

framework

The framework is expected to be able to support the

following functions

Function of request A requester to raise a distributed transaction request

and (after processing) get the result.

Function of

distributed

The coordinator and participants to conduct the actions

of distributed transaction prepare, abort and commit.

57 / 206

transaction

coordinator and

participants

Scalability and

expandability

The framework is expected to be able to support

a, Multiple transaction requests processing in parallel

b, The type of transaction resource being processed in

the transaction can be easily expanded

c, One transaction can contain multiple sub transactions

Ability to handle

network issues

The framework is expected to be able to handle various

network issues(message delay, message lost, etc)

Ability to handle

node failures

The framework is expected to be able to handle node

failure(or node isolation) and recovery during the

processing of transaction

Performance The average time consumed for one transaction(three

resource operations on multiple nodes) in a PCIE based

cluster(32 nodes): <1ms

The average time consumed for one transaction(three

resource operations on multiple nodes) in a Ethernet

based cluster(more than 100 nodes): <15ms

The max number of transactions being handled in

parallel on one node: 20k

The CPU usage is expected to be lower than 1.5%, and

memory usage is expected to be lower than 20M

58 / 206


The requirement analysis document

The architecture document

The design document

The source code

The test environment and the test reports

Two patents

Acceptance Criteria

No. Delivery Acceptance criteria

1

The requirement

analysis

document

The scenarios of requirement are completely

covered. The network topology should be covered

2 The architecture

document

The architecture is expandable and scalable, and

identify all the key technologies

3

The design

document

The design document can support the target

functions. All the key failure/abnormal cases are

well identified and covered

4

The source code The source code is expected to be complete and to

be well verified. If the source code is exploiting the

open source code, the usage should comply with

related Open source library selecting polity of

Huawei

5 Test environment The steps to setup the complete test environment

59 / 206

and test report to verify and measure the framework

The test solution and test report should cover

functionality and performance, the result should

meet the expectation.

The test scenarios should include one of the real

scenarios in storage products.

6 Patents At least two patent ideas and invention disclosure

forms to pass Huawei’s Review

Phased Project Plan

Project

phases

Start End contents targets outputs

Phase

1

Contr

act

Signi

ng

2

months

after

Contra

ct

Signing

1. To reach the

agreement on the

requirement,

delivery and

acceptance criteria

2.Design solution

prototype(including

the open source

library selections)

3.Key algorithm

selection

2.The

acceptance

criteria is

clear

2.To reach

the

agreement

on design

3.To reach

the

agreement

on key

algorithms

1.coorperati

on memo(on

The

requirement

and

acceptance

criteria)

2.The

requirement

document

60 / 206

Phase

2

3

mont

hs

after

Contr

act

Signi

ng

5

months

after

Contra

ct

Signing

1.The architecture

design

2.The design

document(including

key algorithms)

3.patent writing

1.Complete

the

architectur

e

document

2.Complete

the design

document

3.Complete

1 U.S

patent

writing

1.The

architecture

document

2.The design

document

3.1 U.S

patent

document

which is

ready for

filing

Phase

3

6

mont

hs

after

Contr

act

Signi

ng

9

months

after

Contra

ct

Signing

1.patent writing

2.The source code

and test

environment(includi

ng at least one real

scenario in product)

1.

Complete 1

U.S patent

writing

2.

Complete

the source

code

developme

nt

3.

Complete

the

verification

1.1 U.S

patent

document

which is

ready for

filing

2. The

source code

3. The test

solution and

environment

4. The test

report

61 / 206


Subject: Research on Data Placement Algorithm on

Distributed Storage Scene


Background

Distributed Systems Data Placement

For distributed systems, data placement will affect the balance of the entire

storage system, data security, concurrency of storage business, reconstruction

time of disk failure and small migration rate adding disk or reducing disk.

So far, the industry generally uses the CRUSH or consistent hashing algorithm

for distributed systems data placement. But the CRUSH and consistent

hashing algorithm is not fair enough for the load balance in a short time.

So，we need a function for the data placement:

We use three stages to describe the data placement problem:

First, initialization stage, we place the blocks of stripes into the disk.

62 / 206

block block block

block block block

block block block

stripe

stripe

stripe

disk disk disk disk

block block blockstripe

block block block block



Initialization Data Placement

Second, reducing stage, we do the reconstruction, when disk has failure.

disk disk diskDisk

Failure




blockblockblock

Reducing Data Placement

Third, adding stage, we move some blocks from old disks to new disks.

disk disk disk

block block block

block block

block

block

block

block

blockblock

New disk

New disk

block

Adding Data Placement

63 / 206

Distributed Systems Data Visited

We generally use the DHT for the data visited, for example, we use a hash

function for writing/reading a object of one file:

SSD Storage Systems Endurance

Currently, many studies of the flash storage array focus on the write endurance

of solid-state drive only. End-to-end techniques for improving the write

endurance of the flash storage array remain an active area of investigation and

research. In particular we face the challenge of reconstructing the data rapidly

as the solid-state drive wears out.

Flash storage array has occupied some high-end enterprise storage markets,

but the users commonly worry about the write endurance problem of the flash

storage array.

Currently, many studies of the flash storage array focus on the write endurance

of solid-state drive only. End-to-end techniques for improving the write

endurance of the flash storage array remain an active area of investigation and

research. In particular we face the challenge of reconstructing the data rapidly

as the solid-state drive wears out.

Scope

Establish a theoretical model based on the data placed multi-rule

constraints, provide a the theoretical performance bounds, feasible

algorithm and soft implementation

The main constraints include:

A. Security

64 / 206

We make a tree to describe physical units (rack, server, disk) of storage

system, try to place blocks of a stripe at a higher level of physical units (rack

level > server level > disk level) , for example:

There is a rack in storage system, each rack has 7 servers, each server has 20

disks, and each stripe has 3 blocks. So the 3 blocks in a strip are placed into

the different servers is better.

B. Balance of Capacity

C. Balance of specific location

D. Balance of relationship

E. little migration rate

Establish a theoretical model for data placement of heterogeneous.

Consider homogeneous and heterogeneous of medium (HDD/FLASH),

homogeneous and heterogeneous of disk capacity, and homogeneous and

heterogeneous of media interface’s rate of read and write.

Establish a theoretical model for data visited

For random access of the single-stream and multi-stream, how to make the

bandwidth and the utilization of each disk?

Establish a theoretical model for endurance of flash storage system

For SSD, because of its special nature of read and write, we need to do wear

leveling, garbage collection and other operations. So the data placement of

flash storage system should consider these factors to achieve an optimum

solution of global-level and controller-level.

Also, we need to reduce the time of data reconstruction.


The deliverables of the research project include but not limited to the following:

http://www.baidu.com/link?url=9sFm1J48_hDtCoD7sLGB-_XFkocTftM4ZsDMwZXg_uJm3wnkSFThN0aKsOLMb6QBEpdPiwRvQwfddwolLiGH2RETTwmOpvX44i-Jc3iWgTa

http://www.baidu.com/link?url=9sFm1J48_hDtCoD7sLGB-_XFkocTftM4ZsDMwZXg_uJm3wnkSFThN0aKsOLMb6QBEpdPiwRvQwfddwolLiGH2RETTwmOpvX44i-Jc3iWgTa

65 / 206

Survey and analysis report about data placement of homogeneous

physical units, satisfy multiple constraints (security, balance of capacity,

balance of specific location, balance of relationship and little migration

rate).

Survey and analysis report about data placement of heterogeneous

physical units, satisfy multiple constraints (security, balance of capacity,

balance of specific location, balance of relationship and little migration

rate).

Survey and analysis report about data visited about the single-stream and

multi-stream.

Survey and analysis report about endurance of flash storage system.


invention/patents about data placement, data visited, endurance of flash

storage system and data reconstruction of storage system.

Acceptance Criteria

Distributed Systems Data Placement

A. Problem description

M different elements are selected from a set with N elements. During the

selecting process, constraints listed in section 3 should be satisfied with

complexity within ( )O MN ——MN additions and MN multiplications at the

most.

Memory requirement is least than O(N*N).

B. Definition

(1)The collection N elements:

Set with N elements：

app:ds:at%20the%20most

app:ds:at%20the%20most

66 / 206

， ;

(2)Each time we select M( ) elements to format a whole partition：

(4 28)M M different elements are selected from set OSD to format a whole partition ：

，

(3)Define the appearance times of in p times of selections as:

(4) On the basis of (3), the times of a specific location t ( ) is defined

as:

Define the relationship between and after p times of selection as:

For instance,

Hence,

C. Constraints

1、Security

We make a tree to describe physical units (rack, server, disk) of storage

system, try to place blocks of a stripe at a higher level of physical units (rack

level > server level > disk level)

2、Balance of Capacity (The balance of total appearance times)：

，

，

67 / 206

3、Balance of specific location(The balance of appearance times in a specific

location)：

，

，

4、Balance of relationship(The balance of the relationship of a specific osd with

all other osds):

D. Initialization Data Placement

Every time, we select M disks satisfied the constraints 1~4 from the set.

E. Increasing Data Placement

The original partition is:

，

Now we want to increase the number of the candidate set including n elements

fore-mentioned:

， ;

So we will carry on another partition selection, which is the same as the

constraints 1~4, the new partition is:

，

Beyond the above three constraints, there is another constraint to meet:

Define：

st.

F. Reducing data placement

68 / 206

The original partition is:

，

Now we want to reduce the number of the candidate set including n elements

fore-mentioned:

， ;

So we will carry on another partition selection, which is the same as the

constraints 1~4, the new partition is:

，

But, beyond the above three constraints, there is another object to meet:

Define：

st.

Phased Project Plan

One year

69 / 206


Subject: Research on Extra Lower Latency

Interconnection Network


SCM: storage class memory;

RDMA: remote direct memory access;

HPC : high performance computing;

FBCI: fabric based computing infruscture

Background

New fabric for Storage Class Memory Access

Storage class memory (SCM) will be used as IO cache in storage and servers.

SCM has lower access latency, about hundreds of nanoseconds. In fabric

based computing architecture, memory resources will be disaggregated and

distributed in system. Comparing with RDMA fabric latency about 1~2

microseconds, network will be the bottleneck in remote SCM access.

Fig1.

Fig1 shows access latency of each device.

Low latency IO for HPC and FBCI architecture

In 2015, Intel will announce Intel® Omni-Path Architecture. It is designed to

deliver the performance for tomorrow’s high performance computing (HPC)

70 / 206

workloads and the ability to scale to tens—and eventually hundreds—of

thousands of nodes at a cost-competitive price to today’s fabrics.

In the near future, Intel will integrate the Intel® Omni-Path Host Fabric

Interface onto future generations of Intel® Xeon® processors and Intel® Xeon

Phi™ processors to address several key challenges of HPC computing

centers:

1. Performance: Intel Omni-Path architecture will accelerate message

passing interface (MPI) rates for tomorrow’s HPC. Back in June 2014,

when Mellanox debuted the Switch-IB ASICs, it said that the SB7700

provided 7 Tb/sec of aggregate bandwidth across its 36 ports and had

a port-to-port latency of 130 nanoseconds. Dropping the latency in the

Omni-Path gear by 56 percent would put the port-to-port latency down

around an astounding 57 nanoseconds.

2. Cost and density: More components on a server limit density and

increase fabric cost. An integrated fabric controller helps eliminate the

additional costs and required space of discrete cards, enabling higher

server density.

3. Reliability and power: Discrete interface cards consume many watts

of power. Integrated onto the processor, the Intel Omni-Path Host

Fabric Interface will draw less power with fewer discrete components.

71 / 206

Fig.2

Fig. 2 shows Intel Omni Path benefits comparing with current HPC fabric.

Scope

Design, verification and evaluation of:

A low latency IO and architecture for distributed in-memory application in

datacenter;

An architecture, communication protocol and programming model for low

latency remote memory access;

A rack-scale system architecture based on global partitioned virtual address

spaces to provide cost-efficient sharing memory resources;


1. A system prototype and source code (FPGA & Driver) to verify low

latency network;

2. A reliable low latency IO interface design and protocol documents;

3. A verification for benefits of low latency IO in different datacenter

workload;

4. A final report to describe above all;

5. 2 Patents

Acceptance Criteria

The prototype shall include but not limit to the following requirements:

1. Extra low latency ,RTT<500ns @64B;

2. Support high performance distributed memory pool in rack level system;

3. Support compatible with OFED driver;

4. Support torus connection;

72 / 206

5. End to end data integrity;

6. Support electrical and optical interconnection;

Phased Project Plan

1 .5 year

73 / 206


Subject: Research on IO capture and analysis tools

prototype based on the network side in NAS system


Background

There are two purposes of this project:

1. Analyze the workload of NAS in real time and produce performance index.

2. Replay the workload from performance index.

Scope

1. Capture Tools

1) Demands of the capture tool

The tool can capture the data from network and hosts. The hosts should

support Linux and Windows platform. The resource usage of the tool should

not exceed 5% when running on hosts.

i. requirements when capture the data from network

a) The tool can collect the performance index of NAS protocol, such as

requests per second、response time、relationship between request and

response of NFS/CIFS.

b) The tool can collect the performance index of iSCSI、FCoE.

c) The tool can work under heavy load situations, such as high bandwith and

high concurrency. At least 2 IB or 4*10GE throughout is required.

d) The tool should be operated from GUI or command line.

e) The tool should display the captured data in real time.

74 / 206

ii. requirements when capture the data from hosts

a) The tool should capture all the IO operations by timeline, the capture

information of operation should include the file name 、 operation

type(ls/mv/create/open/read/write/delete/append/close/truncate etc) 、file

amount、file size、data size、process id、process name and process

relationship.

b) The tool should be operated from GUI or command line.

c) The tool should display the captured data in real time.

2) Demands of Analysis：

i. The tool should report the following I/O performance indexes: sequential

request count、size distribution、 request interval distribution etc.

ii. The tool should identify workload of some application (application process

and child process)

iii. The tool should support display filters and timeline. The file filter、 process

filter、host filter、time filter、I/O type filter、data size filter are required.

iv. The tool should support the comparison statistic of specified two period.

v. The tool should report the performance index of multiple host in the same

time.

vi. The tool should be operated from GUI or command line.

vii. The tool should support workload replay from captured data.

2. The generation and replay of the session progress

The tool should support workload replay from some configuration file which

produced by analyze.

75 / 206


1. The implementation documents should include total effects of the tool,

especially the effect to the performance, and how to capture performance

index under heavy load.

2. The tool should deployed on server or client host, and should not capture

user privacy data.

3. The implementation documents of the tool should described the analyze

algorithm.

4. The tool should produce workload feature from captured data and

reproduced the workload from the feature.

Acceptance Criteria

1. The acquisition tool could collect the characteristics of data sending to

array completely under the condition of a large number of I/O data

concurrency in the laboratory.

2. The analysis tool could analyze the data collected.

3. The replaying tool could replay IO data completely.

Phased Project Plan

1 year

76 / 206


Subject: Research on 3D TLC NAND Flash Reliability


3D: 3-Dimensional

SLC: Single-Level Cell

MLC: Multi-Level Cell

TLC: Triple-Level Cell

Background

3D TLC NAND flash is emerging as a sustainable storage technology. It

combines 3D integration and TLC flash cells together to achieve high density

and thus brings down prices of flash-based products. Compared to SLC and

MLC flash, TLC flash stores three bits in a single cell and provides higher

density. 3D integration technology stacks multiple layers vertically in a single

chip to further increase density over TLC flash cells. Manufactures such as

Samsung have started mass production of 3D TLC NAND flash. It is promising

to be used in a broad range of applications which require high storage capacity

and low cost, e.g., cold storage.

Although 3D TLC NAND flash has density and cost benefits, it also brings new

challenges due to different storage mechanisms. First, its endurance and

reliability is exacerbated relative MLC flash. Second, its read and write

accesses are slower than those of MLC flash. Third, 3D integration combined

with TLC may introduce new reliability problems. These challenges should be

carefully handled before 3D TLC NAND flash can be used more maturely in

practical applications. Therefore, it is highly necessary to devise novel

methods to optimize both reliability and performance of 3D TLC NAND flash.

77 / 206

Scope



of 3D TLC NAND flash.

2. Establishment of error models for 3D TLC NAND flash.

3. Novel cost-effective wear leveling methods to improve endurance and ECC

algorithms for high reliability.

4. Joint reliability and access performance optimization for 3D TLC NAND

flash.


1. Two patents;

2. One paper;

3. One technical report with respect to the design and implementation of

reliability and performance optimization methods;


Acceptance Criteria


Pass:


1 paper is accepted at a CCF B+ or IEEE flagship conference/transaction AND


Excellent:


78 / 206

More than one paper is accepted by CCF B+ or IEEE flagship

conferences/transactions.

Phased Project Plan

Phase 1: T+4, deliver a patent and a survey on 3D TLC NAND flash;

Phase 2: T+8, deliver a patent;

Phase 3: T+12, deliver one paper, one technical report and simulation/test

results.

79 / 206


Subject: Research on SCM based Data mirroring and

remote replication


SCM: Storage-Class Memory

NVM: Non-Volatile Memory

Background

SCM memories especially NVMs provide high performance just like DRAM, as

well as persistence and high density like storage. They can be used to build a

high-performance, low-latency storage system. Besides, in large-scale data

centers, reliability and availability are critical in the face of hardware, software,

and network failures. However, the state-of-art reliability and availability

approaches for distributed storage systems are not aware of SCMs, but only

considering low-performance storage mediums such as flash and disk.

Therefore, designing a reliable and high available SCM-based distributed

storage system is necessary.

Mirroring and replicating are the commonly used approaches to provide data

reliability and availability. The main challenge should be tracked in this project

is how to maintain data synchronization among multiple mirrors and

replications which are distributed in different SCM-based storage nodes,

without seriously degrading the IO performance that SCMs can support. The

objective is to propose an efficient data reliability and availability mechanism in

the distributed SCM-based storage system.

Scope

The scope of this project includes but not limited to

Data reliability and availability algorithms for SCM

80 / 206

Data mirroring and data replication protocol for SCM

Data mirroring and data replication architecture for SCM

Programming model for SCM


One paper and one patent

A report describing the detailed design and implementation of the

proposed method

Acceptance Criteria

Pass:

1 paper is accepted at the top international conference in the storage area

such as FAST.

1 patent passes Huawei’s review

Fail: Cannot deliver a patent or a paper.

Excellent:

One or more patents are delivered, AND

One or more paper is accepted.

Phased Project Plan

Phase 1: T+6, to deliver a patent

Phase 2: T+12, to deliver a paper

81 / 206


Subject: LUT-based low resource LDPC solution

List of Abbreviations: LDPC, LLR, update, low resource

Background

Low Density Parity Check (LDPC) is a great invention of 20th century, and it is

widely used in flash/SSD based storage for its very high error correction

performance and high parallel characteristics. But its high resource

requirement implemented in hardware is not suited for end-to-end device, or

even for future SSD ECC solution. The objective is to design ways to represent

the LLR in a more valid way and a look-up-table (LUT) based LLR updating

rule, which reducing the logic resource of NMSA by 15% under the situation

that the error correction performance is no worse, or even better.

Scope

The principle to generate original LLR and LUT, the way to optimize the LUT.

Expected Outcome and Deliverables:

1) C code of the new NMSA decoder;

2) documents on LLR generation and LUT generate ways;

3) Simulation results report with certain given code rate, and its comparison

with normal NMSA;

4) decoder in verilog form.

5) Resource analysis report

Acceptance Criteria：

1). We can use the decoder;

2). The LLR generation and LUT generation/optimization principle is

acceptable by the specialists;

82 / 206

3). Error correction performance in the test report shows that the updated

methods achieve almost the same error correction capability;

4). Resource assessment is reasonable and achieved 15% gain.

Phased Project Plan：

2015.9.1-2016.2.28

83 / 206


Subject: The Power Loss Character of Nand when

programming and data recovery

List of Abbreviations:

POR, Data Recovery

Background

When Nand is in programming state, there is an abnormal power down, How

we should recognize the power loss page when power up again and how we

should use the power loss block. We want the analysis about the 2D MLC/3D

MLC/3D TLC Nand power loss character

Scope

To recovery data when EMMC or Nand encounter an abnormal power down as

soon as possible


The Nand Power Loss Character Module setup and Analysis report; Nand

Power Loss Test tools

Acceptance Criteria

1) We can use the analysis to recognize the power loss page and do data

recovery correctly.

2)We can use the Power loss test tools to simulate and test abnormal power

down situation

Phased Project Plan:

2015.9.1-2016.2.28

84 / 206


Subject: ReRAM switching mechanism study and

MLC/TLC feasibility


ReRAM, NAND, MLC, TLC

Background

ReRAM is believe to be the best candidate to replace NAND flash, due to its

potential of 3D stacking capability. It is also been reported that ReRAM can

have MLC operation similar to NAND to increase its density further. However, it

is still unclear what is the best material and what is the best switching

mechanism for ReRAM. There are multiple development projects from each

individual memory vendor involve different schemes.

Scope

To have detail investigation of ReRAM switching material, its switching

mechanism, and MLC/TLC capability of each material.


A report of 1) Possible ReRAM switching materials and switching mechanism;

2) Analysis of each material’s MLC/TLC capability. 3)Pros and Cons of each

category of materials and possible winner.

Acceptance Criteria：

1). Enable us to have a clear understanding of whole ReRAM development

picture.

2)We can use it as a preparation for the possible future technology transition.


2015.9.1-2016.2.28

85 / 206


Subject: UFS hardware acceleration performance

improvement solution


UFS, architecture, performance, hardware acceleration

Background

UFS2.0 is a JEDEC standard to improve mobile devices performance.

UFS2.0 requires high performance and low power consumption to be suitable

for mobile device and wearable device. Hardware acceleration can be used

to handle NAND flash interface command sequences. It can be used to

improve performance and reduce controller resource. However, the limitation

of hardware acceleration is that it is not flexible. There is a need to find out a

hardware acceleration solution in UFS hardware architecture to handle NAND

flash interface command sequences.

Scope

Hardware acceleration solution in UFS to improve performance and reduce

power consumption


1) Hardware acceleration methods in UFS

2) Simulation results on hardware acceleration methods

3) Documents to descript the advantage and disadvantage on each methods

4) Suitable report on ONFI and toggle devices

Acceptance Criteria

1) Hardware acceleration method is suitable in UFS

86 / 206

2) Have been studied all the limitation in hardware and software Hardware

3) Have been studied all the limitation in both ONFI and toggle NAND flash

4) Simulation result and analysis


2015.9.1-2016.2.28

87 / 206


Subject: UFS firmware architecture solution


UFS, architecture, firmware

Background

UFS2.0 is a JEDEC standard to target to replace eMMC 5.1 for mobile devices.

UFS2.0 standard has two additional features to improve the performance.

One is that UFS2.0 standard support CMDq. Second, it has two full channels

for program and read. In order to optimize the best performance, UFS

firmware architecture need to be studied.

Scope

Firmware architecture solution in UFS devices


1) Firmware architecture solution in UFS devices

2) Firmware architecture solutions simulation and result

3) Firmware modules debug methods

4) System performance improvement

Acceptance Criteria

1) Firmware architecture solution is suitable for UFS

2) Have been studied all the firmware architectures solutions

3) CMDq in firmware architectures

4) Firmware architecture can be implemented

88 / 206


2015.9.1-2016.2.28

89 / 206

Computing Technology（HIRPO201505）


Subject: Research on Hadoop & MPP DB using

ServerSAN to accelerate performance and space

efficiency in virtualized environment


Background

Bigdata and MPP DB now use commodity x86 servers as both computing and

storage hardware widely, but in virtualized cloud environment, when integrate

with infrastructure layer, they may be not very efficiency.

For example, in Amazon EC2, people build hadoop on top of EBS. At EBS

layer, the data is distributed with 3 copies, and in hadoop HDFS layer, the data

is also distributed with 3 copies, so there will be 9 copies for application data in

hadoop on EC2, that’s not efficient. This will also happen in MPP DB with

EBS. As shown in Figure 1, the data blocks are replicated across the

DataNodes at the HDFS layer and then each data block will be replicated

across the StorageNodes again at the EBS layer.

Another problem is data locality problems. In EBS, the data is distributed

across different servers and don’t have any locality. If hadoop program is

running on top of EBS, they will also don’t have any locality, so the

performance could be worse than hadoop on bare metal physical hardware. As

shown in Figure 1, when the client read data from Server1, although data is

logically located at Server1, but physically, it is on different servers on storage

layer. So, the applications lose data locality and performance potentially will

suffer.

90 / 206

Huawei’s FusionStorage is a ServerSAN product, and the data stored in

FusionStorage is distributed across the cluster using 3 copy or some kinds of

Erasure Code. When integrate with hadoop or MPP DB, the data is also copied

in HDFS or MPP DB layer, and don’t’ have locality properties. So, we want to

research to solve these two inefficient problems.

Scope

Solve the two problems happened when ServerSAN meet Hadoop or MPP

DB:

One is for storage efficiency, find some way to solve it and make storage

more efficient than original Hadoop on physical Hardware.

Another is data locality, find some way to solve it and make performance to

be higher than original hadoop on physical hardware.


The design documents for how to improve the performance and space

efficiency when ServerSAN meet Hadoop or MPP DB.

91 / 206

A prototype system to show the improvement of the performance and

space efficiency when ServerSAN meet Hadoop or MPP DB.

The benchmark result of performance and space efficiency of the

prototype system.

Some code pieces of the prototype system.

It will be better if produce some patents

Acceptance Criteria

When Hadoop or MPP DB uses ServerSAN as storage infrastructure, the

storage efficiency is not less than using physical Hardware.

When Hadoop or MPP DB use ServerSAN as storage infrastructure, the

performance is not less than using physical Hardware.

Phased Project Plan

Project

Phase Start Time End Time Content Objective Output

Phase

1

Contract

signing

End of the

fourth

month

after

contract

signing

Improve the

space

efficiency

when Hadoop

or MPP DB

use

ServerSAN

as storage

infrastructure

When Hadoop or

MPP DB use

ServerSAN as

storage

infrastructure, the

storage efficiency is

not less than using

physical Hardware.

Design

documents,

Prototype

systems,

Benchmark

result of

space

efficiency of

the

prototype

system

92 / 206

Phase

2

Beginning

of the fifth

month

after

contract

signing

End of the

twelfth

month

after

contract

signing

Improve the

performance

when Hadoop

or MPP DB

use

ServerSAN

as storage

infrastructure

When Hadoop or

MPP DB use

ServerSAN as

storage

infrastructure, the

performance is not

less than using

physical Hardware.

Design

documents,

Prototype

systems,

Benchmark

result of

performance

of the

prototype

system

93 / 206


Subject: Research on Performance Improvement for

Open vSwitch Forwarding


DPDK: Intel’s Data Plane Development Kit

IaaS: Infrastructure as a Service

NFV: Network Function Virtualization

OVS: Open vSwitch

SDN: Software Defined Network

Background

Open vSwitch (OVS) is one of important technology to enable Software

Defined Networking (SDN) architectures and is widely used in IaaS cloud

environment. The performance of OVS is very crucial since it is responsible for

forwarding network traffic between any VMs running in cloud environment. . As

the IaaS Platform scales and emerging Network Functions Virtualization (NFV)

technology, the number of flows and the bandwidth demands increase

dramatically, so there is growing challenge for the performance of OVS.

Community has been already working on and identified some performance

bottlenecks of OVS. However, we believe there is a further work still required

to meet the performance requirements. For instance, the user space switching

solutions leveraging Intel’s Data Plane Development Kit (DPDK) and OVS is

one of concerns, since they also run in virtualization platform just like normal

OVS solutions.

Therefore, user space switching solution which supports at least one of the

three mainstream virtualization technology (VMware, KVM and XEN) and

94 / 206

provides high forwarding performance is meaningful for IaaS cloud

environment. This will be the main focus of this research.

Scope

User space switching prototype with high forwarding performance

This study shall focus on providing a user space switching prototype which

supports at least one of the three mainstream virtualization platforms (VMware,

KVM and XEN), and the forwarding performance needs to achieve the

expected targets shown in section 7.


We expect the outcome and deliverables as following:

Implementation and related design documents of user space switching

prototype

Validation and test reports for the optimization measurement

Acceptance Criteria

The Tn-v testing model is the universal testing model which expects to get the

switch’s forwarding capability under fixed packet loss rate:

95 / 206

We expect performance targets of the switching prototype under Tn-v testing

model (2.5 GHZ CPU, 1e-8 packet loss rate) as following:

Packet Length Current(Kpps) Target(Kpps)

64 1150 >=3100

512 820 >=2300

1518 789 >=810

Phased Project Plan

Project


Phase

1

Contract

signing

End of the

eighth month

after contract

signing

Design and

implementati

on of user

space

switching

prototype

A prototype of

user space

switching

solution

Design

documents

and

implementatio

n

96 / 206

Project


Phase

2

Begin of the

ninth month

after

contract

signing

End of the

twelfth month

after contract

signing

Performance

optimization

of user space

switching

prototype

Make the

solution to

achieve

performance

targets

The final

implementatio

n and

performance

test reports

97 / 206


Subject: Efficient equation rating engine


Background

In recent years, Along with development of communication, the carrier provide

abundant and various services for the customers. For example, the customer

can use voice call, sms, video,data service, etc. And the carrier can provide a

package of some services. All the service need be charged based on the

service type and the customer's usage. Different service has different charge

model. And in the future, the new service will be provides. And the wide band

will develops, and the latency of service need be very small. To support so

many customers use so many services, the performance will be very

importance.

Currently, the system rate for the customer's consumption, every service’s can

has its condition, calculating model. To support all the service, the developers

need write logic for each service. When the more services need support, the

logic will be more complex, and the performance will degrade more.

Scope

To support the variable services and the future development, we need find a

good algorithm to improve the performance:

1）Build a model for the complex rating features. The model shall match the

current rating features, and need easily to expend.

2) Implement the algorithm to process the features. And the algorithm shall be

integrated easily.

3) The performance need increase 50% at least than current, and the memory

space shall not increase than 50%.

98 / 206


Deliverables Description

Design

document

Including but not limited to：

1:Core algorithm and Service Model

2:Main modules and their correlation

3:Key data structures

4:Definition of application interface

5:Re-development manual

Source code of

demo program

1:Source code should be C/C++ language that can pass through

the GCC compiler.

2:The demo program can execute on SUSE11SP1 and

SUSE11SP3.

3:The demo can be used to test cases.

4:Provide APIs for re-development.

Function test

report

Huawei will provide the test cases, the report should describes

each cases’s result.

Performance

test report

Huawei will provide the performance test model and test cases.

And Huawei will re-develop based the API provides by the

co-worker to execute the test cases. The performance test

report shall contain the system resource used, the latency,ect.

Patent analysis

report

1:Analysis on whether or not the solution has infringed any current

patents.

2:If the solution can educe some patents, they shall be described in the

report.

99 / 206

Acceptance Criteria

Deliverables Acceptance Criteria

Design document 1:The model can match all scenario

Huawei list.

2: Describe the algorithms theory.

Source code of demo program 1:Huawei can easily to use APIs to

re-develop.

2:Huawei provide the performance test

cases, the CAPS need increase 50% at

least than current system.

Phased Project Plan

Project

Phase

Start

Date End Date Objects Deliverables

Phase I T T+2months

research

prepare of

efficient

equation rating

engine

1.requirements specification of

efficient equation rating engine

2.Model design

3: Algorithm discuss and comfirm.

Phase II T+2 T+3.5

months Coding

1.APIs design

2.Algorithm implement

3.Simple user cases test.

Phase

III

T+3.5

months T+6months

Project

approval

1:Test the function user cases;

2:Test the performance user cases

3:Improve the algorithm and codes.

100 / 206


Subject: Lane level local path planning


RSU Road Side Unit

OBU On board Unit

CSU Central Service Unit

ITS Intelligent Transport System

Background

With the development of smart transportation, the traditional path planning has

been unable to meet the requirements, and the lane level path planning is the

inevitable trend of future development.

According to the architecture of ITS, the roadside unit (RSU) is deployed along

the road, this project through the RSU equipment for lane level path planning,

aiming at several typical scene design reasonable and effective algorithm for

traffic path planning and scheduling, making road resources be fully used, the

overall traffic efficiency maximization.

Scope

1. Local path planning and scheduling algorithm for multi Lane straight line

Design a lane level local path planning algorithm, to achieve local path

planning for a multi Lane roads (including go straight, turn left, turn right,

complex lane), the design to achieve the following results：

1) Driving safety and reliability, the probability of accidents due to the

algorithm is 0.

2) The full use of lane resources, load balance, in the lane saturation reached

30%, the lane load deviation is less than 20%.

101 / 206

3) Less lane change, and there is no lane change of "A-B, B-A".

4) The scheme is general and can adapt to different numbers of lanes

2. Local path planning and scheduling algorithm for road exit, entrance


planning for the exit and entrance of the highway or urban road, the design to

achieve the following results:

1) Exit and entrance are safety and reliability, the probability of accidents due

to the algorithm is 0.

2) Less impact on other vehicles, and the speed of the affected vehicles is

less than 30%.

3. Local path planning and scheduling algorithm for traffic intersection


planning for traffic intersection(intersection, irregular intersection), the design

to achieve the following results:

1) The intersection traffic is safety and reliability, the probability of accidents

due to the algorithm is 0.

2) The efficiency of intersection is maximized, and the space utilization rate of

the road is reached 100%.

4. Handover algorithm of two local path

During the running process of the vehicle, because the local path is planned by

multi RSU, then, when RSU handover, the two local path will be handover also.

Therefore, the design of a local path handover algorithm, achieve reasonable

handover of the two local path, the design to achieve the following results:

1) The handover is safety and reliability, the probability of accidents due to

the algorithm is 0.

102 / 206

2) No lane change can occur when handover.

3) Before and after the handover, the path is overall reasonable, not local

reasonable.

The above algorithm, are required to have the generality, for example, can be

used for different number of lanes, can be used for different forms of

intersection and can be used for different turning lane and complex Lane.


Delivery the algorithm of local path planning for each scenario.

Acceptance Criteria

1) The algorithm meet the needs of the scene, and can be implemented.

2) The algorithm has been reviewed by experts.

Phased Project Plan

1) T-T+3 (month)： Requirement clarification and communication.

2) T+3-T+9 (month)： Algorithm design.

3) T+9-T+12(month)： Review and acceptance.

103 / 206


Subject: Scalability and Parallel Features of

Algorithms and Its Applications in Matrix Computation


BLAS: Basic Linear Algebra Subprograms.

LAPACK: Linear Algebra PACKage.

PBLAS: Parallel Basic Linear Algebra Subprograms.

ScaLAPACK: Scalable Linear Algebra PACKage.

MKL: Math Kernel Library.

SVD: Singular Value Decomposition.

Background

As we all know, we already enter into the multi-core, many-core and cluster

computing era. On the one hand, in order to fully utilize these computing

resources, conventional serial algorithms are needed to be parallelized. On the

other hand, due to the low computing efficiency, conventional serial algorithms

are hard to exert their effect when dealing with large volume of big data. It

makes sense to investigate scalability and parallel features of algorithms so as

to achieve maximal parallelization degree of algorithms.

Matrix computation, such as matrix-matrix or matrix-vector multiplication,

matrix factorization, calculation of inversion of matrix, and so on, is the basic

operations of many algorithms, including graph algorithms, machine learning

algorithms etc. The performance of matrix computation is essential to the

performance of those algorithms.

104 / 206

In high performance computing domain, several high performance matrix

computation libraries have been developed, such as BLAS, LAPACK, PBLAS,

ScaLAPACK, which are developed in C, C++ or Fortran. However, for big data

or enterprise applications, the ecosystem is in higher level languages such as

Java or Scala. We need high performance matrix computation library in Java

or Scala. Although some Java or Scala matrix libraries have been developed

too, such as Netlib-java and Breeze, they are a wrapper of low level BLAS and

LAPACK. The most popular big data platform now can call these libraries, and

even further accelerate matrix computation by calling Intel MKL. However, the

performance of these libraries is still not good enough for big data applications,

especially on Spark, which is the so-called next generation of big data

analytical platform.

On the whole, this project aims to investigate scalability and parallel features of

algorithms, and further combine the specific domain, that is, matrix

computation, to develop high performance and high scalability of parallel

matrix-related algorithms, such as matrix-matrix multiplication, matrix

factorization (for example, SVD decomposition), calculation of inversion of

matrix for big data applications.

Scope

Based on the above background information, by using matrix computation as

the basis, the scope of this project includes but not limited to:

Analyze the machine-independent properties and structures of algorithms

from both theoretical and implementation points of view.

Analyze information graph and describe the resource parallelism of

algorithms.

Analyze and describe data locality and computation locality.

105 / 206

Analyze dynamic characteristics and efficiency of algorithm

implementations.

Analyze scalability of algorithms, propose a parallel scalability

model/method to describe the scalability of parallel algorithms when the

system scale changes and problem size changes, investigate the effect of

system scale and problem size on the scalability, find which factors

preventing the achievement of high scalability, and further put forward an

effective way to improve the scalability.

Put forward possible methods and considerations for parallel

implementation of the algorithms, and further guide to develop high

performance and high scalability of parallel matrix-related algorithms, such

as matrix-matrix multiplication, matrix factorization (for example, SVD

decomposition), calculation of inversion of matrix for big data applications,

etc. Here, the size of matrix should attain 50 million * 50 million, and in

most cases, the matrices are sparse. Optionally, the matrix computation is

suggested to run on Spark platform.


1. One patent;

2. Parallel matrix computation implementation code and simulation/test results;

3. One detailed high-quality design/analysis/test report with respect to the

theoretical analysis and implementation of parallel algorithms.

Acceptance Criteria

Fail:No patent / parallel matrix computation implementation code is delivered

Pass:

Algorithm analysis and implementation of matrix computation with 3 times

performance gain or above when compared to the benchmark AND

106 / 206

1 detailed high-quality design/analysis/test report

Excellent:

1 patents or above are delivered OR

1 paper is accepted at a CCF B+ or IEEE flagship conference/transaction OR

Algorithm performance gain of matrix computation attains 5 times or above

when compared to the benchmark

Phased Project Plan

Phase 1: T+6, one technical report, and possible patent or paper ideas.

Phase 2: T+12, deliver parallel matrix computation implementation code and

simulation/test results and analysis.

107 / 206


Subject:Researches on Deep Learning

based Chinese Image Text Recognition


Background

Chinese image text recognition has been regarded as an important and

long-standing task in cognitive computing. The importance of this

study even has witnessed in the era of mobile

internet, where both capturing scenes with text and inserting text on

images are much easier and sharing them is becoming an increasingly hot

trend. With a large amount of image texts created every day, however,

automatic image text recognition is considered as a challenging task, as there

are thousands of different text categories. Moreover, the text may with low

resolution, superimposed in complex background, and with large distortion due

to changes of capturing lights, viewpoints, etc. Accurate and

robust Chinese image text recognition is still a task not well

solved. Nevertheless, the value of the task is highlighted in many applications

like image understanding, information retrieval, and internet monitoring.

With the inspiration of advancements of using Deep learning techniques

in solving challenging visual tasks, the objective of this project aims at

developing accurate and robust deep learning based techniques for Chinese

image text recognition. The research outputs are expected

to have a positive effect in above mentioned as well as other related

applications, especially in Chinese dominated environments.

Scope

The research scope is to develop accurate and robust Chinese image text

recognition methods based on deep learning techniques, filling the gap of

108 / 206

cognitive computing from the angle of text recognition. Image text recognition

is a task emerging in the past a few years.

Currently, most existing methods focus on studying English text,

and they basically follow a traditional pipeline that is tailored for text

recognition in scanned documents, in which the text often has little view

distortion and with high resolution, and the backgrounds relatively clean. It is

obvious that image text hardly coincides with these assumptions. On the other

hand, other methods take (some of) the assumptions into account, but they

are only validated on a few datasets with fixed scenes and limited characters,

which might not be well generalized to practical situations.

The project aims at addressing the above shortcomings and proposing

novel solutions for Chinese image text recognition by leveraging deep learning

techniques. Specifically, the problems to be solved are twofold:

1) Large-scale Chinese image character recognition.

Image character recognition aims at recognizing one character individually

from a character (sub-) image. It is a critical preliminary step for image text

recognition. The problem to be solved in this step is to develop a practical

recognition model for Chinese characters. Here, the term practical refers to

that the model should recognize at least 3885 basic characters, which

includes 3755 Chinese primary characters, 26 English characters, 10 digits,

and 94 punctuations from the ASCII table. The average minimum

recognition rate requirement should be no less than the state-of-art one.

2) Chinese Image text recognition

Image text recognition takes a line of characters (i.e., textline) as input and

output a recognized phrase or sentence. In this step, we demand accurate and

robust algorithms that not only support textlines with different styles (i.e., color,

font, degree of thickness) and resolutions, but also can resist to different levels

109 / 206

of noises (e.g., background with different complexity). The average minimum

recognition rate requirement should be no less than the state-of-art one.

3) Create Dataset with Complex Image Background

Extract textline from news images. For example,


1 paper.

1 academic research report.

Acceptance Criteria

(1) Pass:

1 paper is accepted by a conference or journal ranked in CCF B+, or without

rejection in the first review.

1 academic research report includes detailed performance evaluation for key

technologies compared with available equivalent system.

(2) Fail: cannot deliver 1 paper, and 1 academic research report.

(3) Excellent:

More papers are delivered, or one paper is accepted by a conference or

journal ranked in CCF A; or one patent is delivered.

Phased Project Plan

1 year.

Phase 1: T+6, to deliver a paper.

110 / 206

Phase 2: T+12, to deliver 1 academic research report.

111 / 206


Subject: Analysis and design of programming models

of neuromorphic systems with applications on spiking

deep neural nets


SNN: spiking neural networks

CNN: convolution neural networks

DBN: deep belief network

GPU: graphic processing unit

Background

Deep learning has achieved enormous success in a variety of applications

such as speech recognition, computer vision, semantic analysis, and so on.

Deep learning algorithms achieved the state-of-the-art performance on many

benchmark problems. However, this has come at a cost of huge computational

burden. As most deep learning architectures such as DBN and CNN require

iterative operations, their implementation on CPUs/GPUs cannot be

sufficiently efficient, especially when energy consumption is taken into

account.

In parallel to deep learning, research on spiking neural networks has received

much attention due to its energy efficiency, scalability, and biological

plausibility. Via sequences of spikes and their precise timing, SNNs are

capable of encode information in both spatial and temporal domain. More

importantly, SNNs offer insights into human brains, which shed a light for

strong AI in near future. It would be an inspiring idea to combine SNNs with

deep learning architectures as it has a great potential to provide a fast, efficient,

and scable learning framework.

112 / 206

Several challenges shall be tackled in this project: First, how to encode the

learning rules of deep learning algorithms using spiking representation?

Secondly, how to explore the characteristics of SNNs such as firing rate to

improve the learning capabilities? Thirdly, how to develop a unified adaptive

neural coding and learning structure for various applications?

To tackle these challenges, it is necessary and rewarding to develop novel

programming models as the traditional sequential programming model of the

von Neumann architecture is unsuited for the neuromorphic systems. New

programming model of neuromorphic systems needs to permit execution of

complex cognitive algorithms and applications while being efficient for the

chips hardware architecture and effective for programmer productivity.

Architectures and programs are closely intertwined and a new architecture

necessitates a new programming model. Development and improvement of

the programming model can promote the application of neuromorphic chips.

No one has yet presented a concrete study or analysis on the key technology

of the programming model of neuromorphic chips. Although IBM has released

its programming model and related software kit of TrueNorth, it is still far from

popularization. What we are eager to know is the difficulty and hot spot of

research on the programming model of neuromophic chips in academia and

industry.

Scope

The scope of this project includes but not limited to

Quantitative/qualititative analysis on the advantages/disadvantages of

spiking deep neural networks compared with conventional deep learning.

Concrete analysis on the programming model of current neuromorphic

chips.

113 / 206

The direction of development for the programming model of neuromorphic

chips.

Some key technologies to break through for the programming model of

non-von Neumann architecture to bring efficiency for the neuromorphic

chips and effectiveness for programmer productivity.

Design of new programming model of neuromorphic chips.

Novel architecture or language for programming model of neuromorphic

chips.

Application on pattern recognition, computer vision, and associative

memory.


One paper

A report: survey on the existing programming models of neuromorphic systems,

bottlenecks, and future directions.

A report: survey on the existing approaches to spiking deep neural networks

including recent progress, state-of-the-art performance, problems to be solved,

and future directions.

A report: detailed design and implementation of the proposed programming

models and its applications on spiking deep neural networks.

Acceptance Criteria

1 paper is accepted at a CCF B+ conference or an IEEE flagship conference

such as IJCNN, IEEESMC, ISCAS, CDC and so on. AND

Three reports pass Huawei’s review.

Phased Project Plan

Phase 1: T+6, to deliver two reports

114 / 206

Phase 2: T+12, to deliver a paper and a report.

115 / 206

BigData Technology（HIRPO201506）


Subject: Redis Cluster Reliability Improvement


Term Meaning

Redis Redis is an open source, BSD licensed, and advanced key-value cache

and store. It is often referred as a data structure server since keys can

contain strings, hashes, lists, sets, sorted sets, bitmaps, and

hyperloglogs.

Background

Currently, the primary/secondary switchover mechanism of a Redis cluster

depends on other available primary nodes in the cluster for election. If the

number of faulty master instances exceeds half of the total number of master

instances in the cluster, the entire cluster fails, providing poor reliability.

The following figure shows a Redis cluster.

Nine Redis instances form a cluster, with three master instances connecting to

two slave instances. When any two master instances are faulty, only one

M

S S

M

S S

M

S S

M: master S: slave

http://redis.io/topics/data-types-intro#strings

http://redis.io/topics/data-types-intro#hashes

http://redis.io/topics/data-types-intro#lists

http://redis.io/topics/data-types-intro#sets

http://redis.io/topics/data-types-intro#sorted-sets

http://redis.io/topics/data-types-intro#bitmaps

http://redis.io/topics/data-types-intro#hyperloglogs

116 / 206

functional master instance exists, which is less than half of the total. In this

case, election cannot be performed, and the slave instance cannot become the

master instance. As a result, the entire cluster fails and cannot provide

services externally.

In addition, the switchover time between a single pair of nodes sometimes is

too long, exceeding 75 seconds (five times of the cluster-node-timeout

value).

Scope

Solve the two problems described above:

Making sure the entire cluster works normally when the number of faulty

master instances exceeds half of total master instances in the cluster.

Decreasing the switchover time between a single pair of master/ slave

nodes.


Design documents and prototype demo of the solution reaching the

acceptance criteria in session 7.

Acceptance Criteria

When consistency is ensured, no matter how many master instances are

faulty, the slave instance can become the master instance as long as the

slave instance is online, and services on the cluster are not affected.

The switchover time between a single pair of nodes is within 20 seconds.

Primary/secondary switchover is not incorrectly performed when service

volume is large.

Phased Project Plan

Project Start Time End Time Content Objective Output

117 / 206

Phase

Phase 1 Contract

signing

End of the

fourth

month

after

contract

signing

Solution

design and

LLD

Solution

design and

LLD

complete.

Solution

design and

LLD

documents

Phase 2

Beginning

of the fifth

month after

contract

signing

End of the

twelfth

month

after

contract

signing

Redis

switchover

optimization

Redis

switchover

is stable

and

reliable.

The

switchover

time

between a

single pair

of nodes is

within 20

seconds.

Prototype

demo

118 / 206


Subject: Hybird Big Data Cloud Services Technology

Research


Term Meaning

Hadoop Hadoop is an open source big data system from Apache Foundation,

contains HDFS, MapReduce, HBase, Yarn core component, and many

eco system component like Hive, Spark and etc.

Background

Currently, PB-level data is generated in the financial, weather forecast,

simulation, and data mining fields. Large number of computing resources is

required to process PB-level data. However, build large datacenter for

enterprise for big data will cost lots of time and money. Another cost efficient

way is leverage public cloud resource.

To adapt this requirement, the following diagram describes the potential way.

Extension mode: The enterprise cluster inside enterprise consider the

public cloud resource as the same as local resource.

Pull mode: The enterprise cluster only leverage the compute resource on

cloud when local resource not enough. The workloads distribute to public

cloud still access the data inside enterprise.

Push mode: Compare to pull mode, the data is full or partial upload to

public cloud, and workload running with data locality.

119 / 206

Scope

Here are the research questions:

Monitor workload and automatically determine when to leverage public

cloud resource.

Cost aware workload scheduling.

Transparent data access across hybrid cloud.

Data locality scheduling across hybrid cloud.


Demo system for hybird big data cloud services.

One or more patents

Acceptance Criteria

The demo system could work for one of the services from MapReduce, Spark,

or Storm. User could use local resource when resource is enough and could

bursting workload to cloud services immediately when required.

One patent draft passed Huawei review.

Internal Hadoop Cluster

vm

Extension mode

vm vm

Pull Mode

vm vm storage

Push Mode

120 / 206

Phased Project Plan

Project


Phase 1 Contract

signing

End of the

fourth

month

after

contract

signing

Technical

analytics

and

Solution

design and

Finish

solution

design.

Solution

design

documents.

Phase 2

Beginning

of the fifth

month after

contract

signing

End of the

twelfth

month

after

contract

signing

Demo

system

implement

The

enterprise

Hadoop

cluster can

use public

cloud

computing

resources

smoothly.

Prototype

demo

121 / 206


Subject: pan-social data analysis


Background

In recent years , all kinds of online social networks, such as Facebook, Twitter,

Wechat, are emerging, and make the entire internet to be more socialization.

These new social networks are increasingly becoming the sustaining social

interaction platforms for the people, and have become the important media for

carrying online lives and ideas of the people. On the other hand, social network

is no longer a stand-alone platform, and become an integrated platform to link

the business and the users. With the rapid development of mobile technology,

location and time properties are added to the user's online social activities to

better understand user behaviors. With more interactions and information

exchanges, the user's preferences, purchase intentions, friend circle

information can be more deeply analysed.

For the wide range of social networking analysis and research, there are a lot

of interesting topics, such as community detection, influence calculation of the

people, information dissemination model analysis, false information and robot

accounts detection, forecasting of the stock market, the election, infectious

diseases and etc.

For social network analysis platforms and systems, not only well-known

research institutions dedicated to research of social network analysis, the

major IT companies have also set up their own social network analysis tools,

such as SNAP system of Stanford university, AutoMap system of Carnegie

Mellon university, Pregel system of Google. But most systems mainly

supported network level analysis such as structure analysis and visualization

of the social network, while less considering the application level analysis of

the social network (such as user behavior analysis, impact analysis, etc.). In

122 / 206

addition, most of the systems need to enhance the capabilities of content

analysis and efficient index to better support information dissemination

analysis in a large-scale social network.

From the perspective of the data, the data of the telecom carriers endows with

social information of the users, such as the traditional SMS data and call data.

Although in recent years by the impact of the internet, lack a part of the link

information of the users, the data of the telecom carriers still have great

commercial value. In addition, the telecom carriers own the user's location

data based on telecom network, this part of the important digital assets

describes the temporal relationship of the users and can be understood as a

link between the users and the geographical locations. In other words, the

telecom carriers have the pan-social data of the users, including links between

people and people, people and goods (refers to all types of digital content),

people and locations. Based on the above data, the user profiles can be

comprehensive described and used to support more humane marketing

activities and customer care.

Scope

The vision of this project is to build the capacity of mining the valuable

information from the user's pan-social data. The data sources considered

include but are not limited to basic communication data, operating data,

location data owned by telecom carriers, as well as third-party social network

data and internet data. The objectives of this project are:

I) build a scalable platform for pan-social data analysis which has the abilities

of the following aspects:

1. analyse the structure, the features, the dissemination path of the social

network graph;

2. social network graph data visualization and index with high efficiency;

123 / 206

3. content and user based analysis;

II) explore the application of social network data including the following

directions :

1. based on telecom carrier's pan- social data (such as user location and time

relations, communications activities, the user behavior on the internet), using

graph analysis method to explore the social relationship in the social networks ,

and apply this social relationship in value-added services(such as video, music,

novels) and products recommendation as well as other kinds of telecom

carrier's operational scenarios;

2. based on telecom carrier's pan- social data (such as user location and time

relations, communications activities, the user behavior on the internet), such

as the networks of relationships between people and people, people and

things, people and locations, according to a small amount of labeled sample

users (such as students, male, engineer, etc.), identify the features of other

majority of users, and then to complement the user profiles.

3. trace the hot topics and monitor public opinions in the social media,

understand the user 's perception of the brands and the products , sum up

valuable information , improve products and services.


Tasks Deliverables Deliverables description

1.pan-social

data analysis

platform

1. pan-social data analysis

platform software；

2. requirements specification

of pan-social data analysis

platform;

3. functions design

provide the software source

code, requirements

specification, functions

design specification

(including the key technical

points and algorithms) of the

pan-social data analysis

124 / 206

specification of pan-social data

analysis platform;

platform；

2.video-oriented

and pan-social

data based

recommendation

and analysis

technology

1. video-oriented and pan-social

data based recommendation

and analysis software；


of video-oriented and pan-social


and analysis;

3. functions design

specification of video-oriented

and pan-social data based

recommendation and analysis;

4. comprehensive research report

of video-oriented and pan-social


and analysis;


code, requirements




points and algorithms) and

comprehensive research report

of video-oriented and

pan-social data based

recommendation and analysis；

3.user profile

oriented and

pan-social data

based customer

feature

recognition

technology

1. user profile oriented and

pan-social data based customer

feature recognition software；


of user profile oriented and


feature recognition;

3. functions design

specification of user profile

oriented and pan-social data


code, requirements









125 / 206

based customer feature

recognition;





4.social media

based brands,

products and hot

topics analysis

technology

1. social media based brands,

products and hot topics analysis

software；


of social media based brands,

products and hot topics analysis;

3. functions design

specification of social media

based brands, products and hot

topics analysis;



products and hot topics analysis;


code, requirements







products and hot topics

analysis;

Acceptance Criteria

Tasks Deliverables Acceptance Criteria

1.pan-social

data analysis

platform

1. pan-social data analysis

platform software；

2. requirements

specification of pan-social

1.the software source code

consistent with the requirements

and functions design

specifications;

2. the software source code has

126 / 206

data analysis platform;

3. functions design

specification of pan-social

data analysis platform;

no serious and deadly problems

and bugs , implements the

functions points defined in the

requirements specification, is in

line with source code acceptance

criteria of Huawei, and pass the

review of Huawei technical

committee;

4.requirements specification and

functional design specification

compliance with Huawei

standards , and pass the review of

Huawei technical committee ;

5. support real-time social data

access of 10GB per second -level

traffic throughput, and support

more than one hundred million

social nodal analysis;

2.video-oriented

and pan-social

data based

recommendation

and analysis

technology

1. video-oriented and


recommendation and analysis

software；

2. requirements

specification of

video-oriented and pan-social





specifications;





127 / 206

and analysis;

3. functions design

specification of

video-oriented and pan-social


and analysis;

4. comprehensive research

report of video-oriented and


recommendation and

analysis;





committee;





Huawei technical committee;

5. achieve industry- leading level

in the related technical areas;

3.user profile

oriented and

pan-social data

based customer

feature

recognition

technology

1. user profile oriented and


customer feature recognition

software；

2. requirements




recognition;

3. functions design




recognition;


report of user profile oriented

128 / 206

and pan-social data based

customer feature recognition;

4.social media

based brands,

products and hot

topics analysis

technology


products and hot topics

analysis software；

2. requirements


based brands, products and

hot topics analysis;

3. functions design


based brands, products and



report of social media based

brands, products and hot

topics analysis;




specifications;









committee;





Huawei technical committee ;

5. the accuracy of social media

based brands, products and hot

topics analysis is greater than

90%;

Phased Project Plan

Projec

t Phase

Start End Objects Deliverables

129 / 206

Date Date

Phase I T T+3months

1. research

prepare for

pan-social data

analysis platform;


pan-social data analysis platform;

2.functions design specification of

pan-social data analysis platform;

Phase II T+3 T+6months

1.pan-social data

analysis

platform；

2.application

research based on

pan-social data;

1.pan-social data analysis platform

software；

2.comprehensive research report of

video-oriented and pan-social data

based recommendation and

analysis;




analysis;

4. comprehensive research report of

user profile oriented and pan-social

data based customer feature

recognition;

5.requirements specification of user

profile oriented and pan-social data

based customer feature recognition;

Phase

III T+6 months T+9months

1.video-oriented

and pan-social

data based

recommendation

1.video-oriented and pan-social data


analysis software；


130 / 206

and analysis

technology；

2.user profile

oriented and

pan-social data

based customer

feature

recognition

technology；



analysis;

3.user profile oriented and pan-social


recognition software；


user profile oriented and pan-social


recognition;

Phase

IV

T+9 months T+12 months

1. social media

based brands,

products and hot

topics analysis

technology

；


products and hot topics analysis

software；

2.requirements specification of social

media based brands, products and



social media based brands, products

and hot topics analysis;

4. comprehensive research report of

social media based brands, products

and hot topics analysis;

131 / 206

Media Technology（HIRPO201507）


Subject: Development of Internet transmission

protocol with high performance in wired and 3G/LTE

network


Background

The transmission of video stream (e.g. IPTV, OTT video) and file on Internet

needs high utilization of network bandwidth. The low bandwidth utilization of

traditional TCP protocol (e.g. TCP new reno) brings frequent lags in video

buffering and file downloading/uploading, which leads to a poor user

experience. It is much worse in the case of 3G/LTE wireless network which has

higher loss rate and a large amount of random loss events. Some TCP

variants, e.g. TCP vegas, TCP Fit, FAST TCP, have improved the adaptability

of TCP and increased efficiency significantly. However, they are heavily

influenced by high delay jitter or loss rate and have not yet achieved the full

bandwidth utilization.

The project objective is to implement an Internet transmission protocol in wired

and 3G/LTE network with near full effective bandwidth utilization. It is not

allowed to bring any changes into the OS kernel in the user side, so it may be

implemented (a) in the TCP protocol stack in server OS kernel (e.g. KO), or (b)

in the application layer of server and user both (e.g. based on UDP protocol).

Pattern (a) is a method of TCP variant and is easy to implement, but is difficult

to achieve full bandwidth utilization. Pattern (b) is recommended because it is

more flexible and has definitely potential to fully utilize the bandwidth, even

though it needs user to install specified application.

132 / 206

Scope

The research scope includes all aspects of a complete transmission protocol,

such as connection building and closing, congestion control, reliability control,

and flow control et al.

The main work is the research on unilateral (Pattern a) or bilateral (Pattern b)

algorithm which is able to fully utilize bandwidth in wired or 3G/LTE network

environment. And the core work is the design of congestion control algorithm.

The challenges include:

1. The transmission rate should be aware of congestion in network in time.

2. The awareness of congestion should be accurate and not be affected by

random loss (3G/LTE) or delay jitter.

3. The maximum utilization of available network bandwidth any time.

4. The relative TCP friendliness, which means reserve certain bandwidth for

TCP flows sharing the network.

In addition, a transmission protocol performance evaluation method is also a

part of the research content, including the evaluation of the protocol efficiency,

fairness, friendliness, stability, and convergence. The evaluations will give a

synthetical score of the target protocol.


Deliverables Description

Protocol design

document


Core algorithm and protocol interaction flow

Main modules and their correlation

Key data structures

133 / 206

Definition of application interface

Source code of

demo program

Source code should be C/C++ language that can

pass through the GCC compiler.

The demo program can execute on SUSE11SP1

and SUSE11SP3.

Protocol test report


Test data in 3G/LTE simulation network

Test data in wired simulation network with high

delay, high lost rate, and high delay jitter.

Test data in real 3G/LTE and wired network

Test data on fairness

Test data on stability.

Test data on convergence

Test data on data reliability.

Patent analysis

report

Analysis on whether or not the protocol has

infringed any current patents.

Analysis on which parts of the protocol can apply

new patents.

Transmission

protocol evaluation

method documents

Including but not limited to:

Analysis on the current development in protocol

evaluation field

Evaluation methods of the protocol efficiency,

fairness, friendliness, stability and convergence.

Synthetical evaluation algorithm

134 / 206

Evaluation report on

the protocol in this

project

Synthetical evaluation of the transmission protocol

developed in this project

Acceptance Criteria

Transmission protocol:

1. The average effective bandwidth utilization rates in the network

environments in Table 1 meet the requirements in Table 2:

Table 1. Network environments

Network type Network description

Metropolitan network

3G (China mobile)

LTE (China mobile)

LAN

National network Internet from Nanjing to Shenzhen

International network

Internet from China to USA

Internet from China to Germany

Table 2. Network models and effective bandwidth utilization rate requirements

Model

Bandwidth

（bps） Delay（ms）

Loss

rate

（%）

average effective

bandwidth

utilization rate

1 2M～100M ≤500ms（10% jitter）

≤5% ≥90%

≤10% ≥85%

2 100M～1G ≤500ms（10% jitter）

≤5% ≥85%

≤10% ≥80%

135 / 206

2. CPU overhead should not be higher than TCP

3. Intra-protocol fairness (Multiple flows equally share the bandwidth)

4. Friendliness with TCP (Reserve some bandwidth for TCP flow, evaluation

method needs further negotiation).

5. Convergence (Evaluation method needs further negotiation).

6. Stability (Evaluation method needs further negotiation).

Transmission protocol evaluation method:

Achieve the state of the art performance in related research field, with the

criteria on the same or similar experiment conditions.

Phased Project Plan

a) Transmission protocol:

Start Date: T months

End Date: T+6 months

Tasks: Research on transmission protocol

Objects: Full bandwidth utilization in wired or 3G/LTE network environment

Deliveries:

1. Protocol design document

2. Source code of demo program

3. Protocol test report

4. Patent analysis report

5. Communication and guidance:

136 / 206

i. Guidance through mail or phone, and resolve the technological problem in

Huawei team. At least 3 face-to-face meeting for communication of

protocol algorithm.

ii. Progress report meeting every 2 weeks

b) Transmission protocol evaluation method:

Start Date: T+6 months

End Date: T+7 months

Tasks: Transmission protocol evaluation method

Objects:

1. Analysis on the current development in protocol evaluation field

2. Research on evaluation methods of the protocol efficiency, fairness,

friendliness, stability and convergence.

3. Guide Huawei Company to build the Transmission protocol evaluation lab

4. Deliveries:

1) Transmission protocol evaluation method documents

2) Evaluation report on the protocol in this project

3) Communication and guidance:

Guidance through mail or phone, and resolve the technological problem in

Huawei team. At least 3 face-to-face meeting for communication of

protocol algorithm.

Progress report meeting every 2 weeks

137 / 206


Subject: Research on screen content codec’s

performance and coding speed


SCC: Screen Content Coding

HEVC: High Efficient Video Coding

Background

Screen content coding mainly for remote sharing and control, one scenario is

remote desktop sharing in a video conference. Because video conference

work on both the wired and wireless network, so a high compression ratio

codec is needed in the scene. It don’t need high image quality, but it can work

well on the loss network.

The other scenario is desktop cloud, it will instead of PC in office. The officers

use terminals to connect the desktop cloud, and they work for long time with

the terminal, so they require as close as possible to the PC experience. It

needs high frame rate, low delay and lossless image.

Scope


1. Propose lossless compression method to improve the compression ratio of

text region in screen content.

2. Research identification method of text regions and image regions, auto

select lossless and loss compression algorithm.

3. Propose fast motion estimation algorithms on screen content sequence.

4. Propose private progressive image coding method, such as JPEG2000,

but faster than it.

138 / 206

5. Propose private high performance lossless image coding method and

lossy image coding method (similar to JPEG-LS), we can control the error

range for each pixel.


1. algorithm design documents;

2. technical reports of the codec’s performance;

3. source code of the screen content coding codec;

Acceptance Criteria

Fail: Not completed the algorithm in the Section 5.

Pass:

The codec’s encoding speed is faster than 1080P@30fps on a core with 3.0

GHz

The codec’s compression ratio is better than BZIP2 in text region. it can

identify the text and image region, and select proper encoding method

adaptively.

Excellent: The codec’s performance is better than HEVC SCC.

Phased Project Plan

Phase 1: T+4, completed the text lossless coding method and the region

identification method;

Phase 2: T+8, completed the progressive image coding method and lossless

image coding method;

Phase 3: T+12,completed the architecture of the codec.

139 / 206


Subject：Digital Photography, Image Enhancement,

Saliency Detection, Composition Optimization

Background

People often take photographs at tourist sites and these pictures usually have

two main elements: a person in the foreground and scenery in the background.

This type of “souvenir photo” is one of the most common photos clicked by

tourists. Unfortunately, the amateur photographers often take unsatisfied

souvenir photos due to the lack of composition knowledge. Although

algorithms that aid a user-photographer in taking a well-composed picture of a

scene exist, few studies have addressed the issue of properly positioning

human subjects in photographs.

We aim to develop a photographic assistance system. Based on the analysis

of background scenes, the system can compute an optimal human standing

position and recommend it to the users to help them take well-composed

souvenir photographs.

Scope

First of all, the desired system gives guidance to users at the moment of taking

a photo, rather than post-processes a captured photo.

The system works as follows. Firstly, the user choose a view of photography.

Secondly, the scenery should be extracted from the input scene. It can be

finished by asking the user to mark the scenery manually or detecting the

salient regions automatically. Thirdly, the system analyzes the scenery and

identifies the optimal standing position for a human subject, dictated by both

the relative size of the subject to a given scene and his or her position within

that scene. Finally, the system shows an icon in the monitor at the optimal

position and recommend it to users.

140 / 206

Generally, the key techniques involved in this project include

c) extracting/segmenting the desired scenery from the input scene;

d) defining composition rules, which can be derived from the traditional

photography experiences and also can be learned from a large-scale

database;

e) computing optimal standing position based on the rules defined in a).

In summary, there may be three major difficulties in developing such a

photographic assistance system:

a) There are too many composition rules, and some of them are internally

inconsistent. Therefore, how to define effective rules and implement them

in program is the biggest challenge of this project.

b) Since the desired system works during the process of photography, the

algorithm employed must be fast enough. In addition, to ensure the

usability, the system can’t require too many complicate interactions.

c) For some complicated scenes which are not suitable to take souvenir

photographs, the system must give the user reasonable and

understandable explanations and avoid generating terrible

recommendation results.


The expected outcome and deliverables contains

a) system demo (C code)，including the implementation of the

recommendation algorithms and the user interface;

b) design documents of the major algorithms, including scenery extraction,

composition rules definition, human position recommendation, etc.

141 / 206

Acceptance Criteria

a) Accuracy—We build a dataset of background scenes of pre-defined typical

scenes and employ several professional photographers to manually mark the

ideal human standing position based on their experience. The marked

positions are considered as ground truth. Then, the recommendation results

generated by the developed system will be compare to the ground truth.

b) Usability—We will employ some volunteers to try the developed system and

investigate the degree of satisfaction.

Phased Project Plan

Phase Description Deliverables

T - T+2 Project design Design documents

T+2 – T+6 Demo develop System demo (C code)

T: Contract signing date

142 / 206


Subject: Research on AVC-HEVC Transcoder Based

on AVC Decoding Information


AVC: Advanced Video Coding

HEVC: High Efficiency Video Coding

Background

HEVC, a new high-efficient compression technology, is becoming more and

more popular in codec industry. However, video streams compressed using

HEVC are unusual on the internet. AVC is still the most widely-used

compression technology in OTT industry. On the other hand, there are more

and more terminal devices which support HEVC decoding. As a result, carriers

would like use AVC-HEVC transcoding technology to decrease the bandwidth,

as well as give terminal users better experience, because of the high

compression feature of HEVC.

The traditional process of AVC-HEVC transcoding is as follows:

1. AVC decoder transforms the AVC video stream into YUV stream.

2. HEVC encoder gets the YUV stream and transform it into HEVC stream,

without using any other information from AVC decoder.

AVC Decoder HEVC EncoderAVC In HEVC OutYUV

Because of the high complexity of HEVC algorithm, a traditional AVC-HEVC

transcoder can consume a large amount of CPU MIPS. AVC-HEVC transcoder

based on AVC decoding information is emerging as a high-speed transcoding

technology. In this improved transcoding process, the HEVC encoder does not

143 / 206

only use the YUV stream, but also use some extra information from the AVC

decoder, to increase its working efficiency.

Scope



of AVC-HEVC transcoding technology based on AVC decoder.

2. Establishment of algorithm models for AVC-HEVC transcoding based on

AVC decoder.


1. Algorithm code;

2. Algorithm design document;

3. Simulation/test results for the algorithm.

Acceptance Criteria

Fail: Not achieve the goal of arithmetic performance.

Pass:

Compare with X265 1.6 medium option, algorithm performance step up 8x; for

video quality, BD-Rate is less than 5%;

Excellent:

Compare with X265 1.6 medium option, algorithm performance step up 10x;

for video quality, BD-Rate is less than 5%;

Phased Project Plan

Phase 1: T~T+4, deliver a version: algorithm performance step up 8x;

Phase 2: T+5~T+12, deliver a version: algorithm performance step up 8x; for

video quality, BD-Rate is less than 5%;

144 / 206

Security Technology（HIRPO201508）


Subject: Searchable Encryption and Cryptographic

Access Control


DBMS Database Management System

ABAC Attribute-based Access Control

SWHE Somewhat Homomorphic Encryption

IBE Identity-based Encryption

ABE Attribute-based Encryption

Background

Cryptography provides strong security guarantees for data in transit and data

at rest, such as confidentiality, integrity and authenticity. However, applying

cryptographic schemes to existing systems may affect the usability of the

systems. For example, encrypting sensitive data before storing in a database

can protect the data at rest. Nevertheless, existing database functions such as

keyword searching or numeric calculation cannot be performed over the

ciphertext. Achieving certain functionality over the ciphertext is an active

research topic. It is a big challenge to strike a balance between the usability

and security.

Searchable encryption is one of the major research topics in academy as well

as industry. Searchable symmetric encryption (Song et al. 2000) allows data

owner to search his own data with some pre-defined keywords. Public key

encryption with keyword search (Boneh et al. 2004) allows keyword search

145 / 206

with multiple data owners. For numeric data, partially homomorphic encryption

schemes such as RSA or Paillier allow only multiplication or addition of the

original data. Although the recent development of fully homomorphic

encryption (Gentry 2009) enables multiple multiplications and additions, it is

not yet practical for large scale deployment. The order preserving encryption

scheme is suitable for the sorting function of database.

CryptDB is a practical searchable encryption system (Popa et al. 2011) for

DBMS, which use a combination of cryptographic schemes such as

deterministic encryption, Paillier encryption, order preserving encryption, etc. It

supports large number of SQL queries over encrypted data. There exists a

trusted proxy on the client side to translate normal query into “encrypted” query

and send it to the server. The encrypted result will also be decrypted by the

proxy. The throughput loss of CryptDB (compared with no encryption) is about

26%. Recently, Microsoft proposed the use of “Always Encrypted” feature in

SQL Server 2016. They also use a trusted proxy on the client side, with the

extra key management function. The supported SQL query over encrypted

data is less than that of CryptDB.

Access control is traditionally managed by system administrator, using

role-based access control. Attribute-based access control (ABAC) is a possible

enhancement to the access control mechanism. By using cryptographic

method, fine-grained access control can be enforced by the data owner

instead of the system administrator. Attribute-based encryption (ABE)

(Sahai-Waters 2005) allows the recipient to decrypt the ciphertext if and only if

his attributes satisfy the access control policy chosen by the data owner.

Functional encryption (Boneh et al. 2011) is a type of public-key encryption in

which possessing a secret key allows one to learn a function of what the

ciphertext is encrypting. These schemes allow more fine-grained access

control by the data owner.

146 / 206

Scope

The project contains two parts: Searchable Encryption and Cryptographic

Access Control. Both parts should be completed in this project.

1. Searchable Encryption

The aim for this part is to build a practical database system which support

searchable encryption. The technique should be applicable to big data

database such as Hadoop or Apache Spark. An efficient prototype is expected

since huge volumes of data are processed. Major functionality of the database

system should be retained. Possible research directions include (but not

limited to): fuzzy keyword search; search for arbitrary Boolean query including

NOT gate; efficient computation using somewhat homomorphic encryption

(SWHE) [Boneh et al. 2013]; efficient order preserving encryption.

Comparison with existing systems such as CryptDB is highly preferred. The

expected throughput loss (compared with no encryption) is less than 20%. The

security model and assumptions must be clearly stated.

2. Cryptographic Access Control

The aim for this part is to design an attribute-based access control protocol

using cryptographic technique, for data stored in the cloud environment. As a

result, the access control policy is controlled by the data owner, but not the

administrator of the cloud server. Possible research directions include (but not

limited to): ciphertext policy ABE; key policy ABE; functional encryption.

The security and the efficiency analysis are expected. A prototype

implementing the access control scheme is preferred. The related paper

should be published in the following conferences or journals:

Conferences: Level 1: Crypto, Eurocrypt, Asiacrypt, CCS, S&P, NDSS,

USENIX, ESORICS, PKC

Level 2: CT-RSA, FC, AsiaCCS, ACNS, SecureComm, TrustComm

147 / 206

Level 3: ACISP, ISPEC, SAC, ISC, ICICS, Inscrypt

Journals: Journal of Cryptology, TDSC, TIFS, TISSEC, JCS, DCC, IJIS


A prototype and a design report are expected for searchable encryption.

A prototype and a design report are expected for cryptographic access

control.

Two patents and two research papers are expected.

A technical report is expected to summarize the related works and the

design principle of the patents and prototypes.

Acceptance Criteria

Prototype for searchable encryption: The prototype should support SQL

queries not less than that supported by CryptDB. The expected throughput

loss of the system (compared with no encryption) is less than 20%.

Prototype for cryptographic access control: It should support dynamically

allocating access right or revoking access right.

The patents should be reviewed by the technical committee by Huawei.

The research papers should be published in the list of suggested

conferences and journals listed above.

The technical report should be reviewed by the technical committee by

Huawei.

Phased Project Plan

Duration: 1.5 year, 3 phases with 6 months each.

Phase 1: Research for searchable encryption and design the algorithms

for searchable encryption. A patent is expected in this phase. The

148 / 206

proposed scheme should support SQL queries not less than that

supported by CryptDB.

Phase 2: Implement searchable encryption and design efficient

computation algorithm using somewhat homomorphic encryption; start

research for cryptographic access control. A prototype for searchable

encryption is expected in this phase. It should have less than 20%

throughput loss when compared with no encryption. A patent and a draft of

the research paper for efficient computation using SWHE are expected in

this phase.

Phase 3: Research and implement the cryptographic access control. The

prototype should support dynamically allocating access right or revoking

access right. A research paper is also expected in this phase.

Other Information

References

[Song et al. 2000] Dawn Xiaodong Song, David Wagner, Adrian Perrig.

Practical Techniques for Searches on Encrypted Data. IEEE Symposium

on Security and Privacy 2000: 44-55.

[Boneh et al. 2004] Dan Boneh, Giovanni Di Crescenzo, Rafail Ostrovsky,

Giuseppe Persiano: Public Key Encryption with Keyword Search.

EUROCRYPT 2004: 506-522.

[Gentry 2009] Craig Gentry: Fully homomorphic encryption using ideal

lattices. STOC 2009: 169-178.

[Popa et al. 2011] Raluca A. Popa, Catherine M. S. Redfield, Nickolai

Zeldovich, Hari Balakrishnan: CryptDB: protecting confidentiality with

encrypted query processing. SOSP 2011: 85-100.

[Sahai-Waters 2005] Amit Sahai, Brent Waters: Fuzzy Identity-Based

Encryption. EUROCRYPT 2005: 457-473.

149 / 206

[Boneh et al. 2011] Dan Boneh, Amit Sahai, Brent Waters: Functional

Encryption: Definitions and Challenges. TCC 2011: 253-273.

[Boneh et al. 2013] Dan Boneh, Craig Gentry, Shai Halevi, Frank Wang,

David J. Wu: Private Database Queries Using Somewhat Homomorphic

Encryption. ACNS 2013: 102-118.

150 / 206


Subject: Lightweight cryptography performance on

resource-constrained devices


IoT: Internet of Things

Background

IoT devices are expected to proliferate significantly in the coming years. Since

each node must be cheap and even disposable due to limited hardware

resources, energy constraints, cost limitations, and low bandwidth, protecting

the communication between resource-constrained nodes is hard, e.g.,

side-channel attacks. IoT security is the area of endeavor concerned with

safeguarding connected devices and networks in the Internet of things.

Majority of modern cryptographic algorithms were designed for desktop/server

environments, many of these algorithms cannot be implemented in the devices

used by IoT.

Lightweight cryptography, a trade-off technique between lightweightness and

security, is a cryptographic algorithm or protocol tailored for implementation in

constrained environments including RFID tags, sensors, contactless smart

cards, health-care devices and so on. It has been a very hot topic for recent

years, driven by the lack of primitives capable to run on devices with very low

computing power, especially, IoT. Previous researchers proposed lots of

lightweight cryptography schemes, however, less performance evaluation

(Metris or benchmark) are given, e.g., code size, memory size, cycle count,

and so on.

Scope

Huawei Shield Lab is seeking proposals to implement sufficient cryptographic

functions on constrained devices. The cooperation project will design flexible

151 / 206

benchmarking framework to evaluate lightweight cryptography in the IoT

context, e.g., resources-constrained sensors devices (low power and limited

battery). The research scope targets to build an appropriate implementation

platform and test a number of representative algorithms, which may belong to,

but not limited to, the following categories:

lightweight block ciphers

lightweight hash functions

lightweight stream ciphers

lightweight one-pass authenticated ciphers


A research survey on the implementations of lightweight cryptography.

A detailed report on the experiment design and experimental results

obtained, together with reasonable analysis, comparison and discussions.

The implementation platform, demo and source codes developed for the

project.

Acceptance Criteria

Project proposal is accepted by the evaluation team, Shield Lab, Huawei.

Project deliverables are accepted by the evaluation team, Shield Lab,

Huawei.

Phased Project Plan

The project is expected to be done in one year which roughly follows the below

time schedule:

Stage 1 (First four months): Survey lightweight cryptography

implementations and plan how to build an implementation platform

necessary for the project.

152 / 206

Stage 2 (Second four months): Build the implementation platform and

perform initial experimental tests of lightweight cryptography on resources

constrained devices.

Stage 3 (Last four months): Conduct detailed testing implementations of

lightweight cryptography on resources constrained devices and provide

scientific analysis.

153 / 206


Subject: Docker Security Research


VM Virtual Machine

Container A virtualization method where the

kernel of an operating system allows

for multiple isolated user space

instances, instead of just one.

Background

The foundation of cloud computing is converged infrastructure and shared

services. Container, as a lightweight virtualization, gains more and more

attention in cloud computing. Because compared to VM, containers cost less

resource, such as memory and disk, while gaining higher performance, such

as booting time. There are many different container resolutions. Among them,

Docker is the most shining star.

Although Docker gains fast development and wide deployment, Docker

security is still an issue. We all have known that Docker or other container is

less secure than VM, but we do not know for sure. How many insecure points?

How to rescue/patch them? And, what is the right place to apply Docker or

other container in cloud computing environment?

Scope

Research Docker and other Linux container security, including classifying

existing Docker security issues, disclosing some unannounced Docker CVEs,

finding resolutions for existing security issues.

The research work shall provide a profile on Docker security. What is the root

cause? What is already discovered? What is not yet exposed? And most

154 / 206

important, the research work shall tell us what should be done on Docker

security.


Obtain a report on Docker security analysis. Get 2 new Docker CVEs. Deliver

a skeleton to fortify Docker.

Acceptance Criteria

The security report shall cover all existing (till the start date of the project) CVE

of Docker and Linux container related sub-systems: namespace, cgroup, and

file systems that Docker uses (AUFS, Btrfs…).

The project shall give us some un-announced CVEs on Docker or Linux Kernel

container related sub-systems.

The project shall generate a skeleton to fortify Docker, with which Docker

container could resist the existing attacks.

Phased Project Plan

The whole project shall last for 1 year, being divided into 3 phases, each

lasting 3-4 months.

1. Analyze Docker security

The first phase is to research the current state of Docker security. In the end of

it, a security analysis report shall be generated, including all Docker, Linux

Kernel namespace, Linux Kernel Cgroup, and Linux AUFS CVEs. The report

shall not only list CVE number but also the description.

2. Explore Docker, get 2 new CVEs

This phase shall generate 2 un-published CVEs, including code to exploit

them.

3. Develop a skeleton to fortify Docker.

155 / 206

The last phase is to develop a prototype, via which Docker can be protected to

existing exploitation and attacks.

156 / 206


Subject: Hardware Device authentication with

hardware fingerprint in M2M of IoT

M2M System Model of IoT

The increasing use of internet of things (IoT) leads to the pressing needs for

device authentication of M2M in wireless networks. Our main goal behind

device authentication is to overcome one of the major limitations of existing

solutions—the inability to effectively detect Medium Access Control address

spoofing. So we explore the use of clock skew of a wireless network access

point as its fingerprint to detect unauthorized APs quickly and accurately.


M2M: Machine to Machine


RSS: Received Signal Strength

CSI: Channel State Information

MIMO: Multi Input Multi Output

OFDM: Orthogonal Frequency-Division Multiplexing

157 / 206

Background

With advances in microtechnology and wireless networks, networked mobile

systems are becoming increasingly prevalent. There is also an ever growing

demand for ubiquitous services. These two factors are fueling a wide-scale

deployment of wireless networks including device authentication and key

establishment.

However, because of their importance in providing ubiquitous services and

their inherent vulnerability due to broadcast nature of the wireless medium, the

wireless networks of IoT are also becoming targets of a variety of attacks. One

of the ways in which a wireless network can be attacked is by introducing one

or more unauthorized fake Access Points (APs) in the network. A fake AP can

be set up by a malicious attacker to masquerade as an authorized AP by

spoofing the authorized AP’s medium access control (MAC) address. This fake

AP is used to fool a wireless node into accessing the network through the fake

AP instead of the authorized one. The fake AP can then launch a variety of

attacks thereby compromising the security of the wireless communication.

Setting up fake APs is not hard. Public domain programs including rglueap

sniff 802.11 probe request frames to find out the default AP of the probing

wireless node, and then, impersonate the default AP. Therefore, device

authentication is a very important task of wireless network.

The new wireless security enhancement RSNA(Robust Security Network

Association) uses traditional cryptographic methods (i.e., digital certificates) to

provide strong mutual authentication between wireless clients and the APs.

Although this solution, if implemented properly, will make the fake AP attack

less likely, the following practical issues can still make wireless networks using

RSNA vulnerable. First, management and verification of digital certificates

across different domains are known to be cumbersome. Second, as the

current AP selection algorithms use signal strength as the only criteria for AP

selection, users can be fooled to connect to the fake AP that has a higher

158 / 206

signal strength compared to the original one but does not support any security

measures such as RSNA.1 Third, an attacker can also set up fake APs having

the same identifiers (MAC address, basic service set identifier (BSSID) and

service set identifier (SSID)) as the original AP and evade detection by using

different physical channel characteristics (by using short/long preambles,

operating in a different channel, etc.). These facts motivate us to find a viable

noncryptographic solution to device authentication. We emphasize that this

solution is not meant to replace existing cryptographic methods. Rather, it

should be used in conjunction with the cryptographic methods to achieve a

higher level of security in wireless networks. The current state-of-the-art

non-crypto methods for unauthorized AP detection cannot detect fake APs.

Scope

We will explore a passive online scheme that can authenticate hardware

device with high accuracy and minimum overhead. This scheme, for

fingerprinting personal computers and servers, is based on estimating clock

skews of hardware devices. An hardware device’s clock skew acts as its

fingerprint.

Bahl et al. proposed a method to detect fake APs and authenticate hardware

device by monitoring the anomaly in the monotonicity of the“sequence number”

field of beacon frames sent by the authorized AP and the fake AP which is

masquerading as the authorized one. However, this method can only detect

the presence of a fake access point; on the contrary, our scheme can detect

and separate out packets from fake AP. Another serious drawback of this

method is that it will only work if both the authorized AP and the fake AP are

active at the same time. Bahl et al. also suggested the use of a location

detection algorithm to detect the fake AP if the authorized AP is inactive at the

time of detection. The accuracy of this method depends on the accuracy of the

location detection algorithm. If the fake AP operates at a location that is very

close to the authorized AP’s working location, then this location detection

159 / 206

method will be ineffective. Our solution removes these constraints and detects

unauthorized APs in realistic scenarios.

Yin et al. proposed a method for detecting rogue APs and authenticate

hardware device that also act as layer 3 routers. However, this work is also

vulnerable to MAC spoofing. Franklin et al. introduced a technique to

fingerprint wireless device drivers. However, an attacker can also use fake APs

with the same wireless device drivers by choosing the same model and the

same manufacturer as the original one to evade detection.

Use of clock skew to fingerprint a remote device is not new. Kohno et al. have

already shown that clock skew can be used as a reliable fingerprint for a

device. However, Kohno’s scheme focused on wide-area wired networks. Its

application in a local area setting can result in higher accuracy. Unlike Kohno’s

scheme that uses TCP/ICMP time stamps, in our scheme, we use the Time

Synchronization Function (TSF) time stamps in the beacon/probe response

messages sent by the AP, to determine its clock skew. Therefore, our

contribution is significant because we apply the clock-skew-based

fingerprinting to a scenario where the detections are much faster, accurate,

and less vulnerable to spoofing attacks compared to Kohno’s original scenario

that uses TCP time stamps.

Use clock skews to be an efficient and robust method for Device

Authentication Device in wireless local area networks.


1patents:

1) Clock skew based device authentication method in M2M communication.

1 prototypal system:

1) Hardware Device authentication with hardware fingerprint in M2M.

160 / 206

1-2 Papers

Acceptance Criteria

Hardware Device authentication success rate 99%.

Phased Project Plan

2015.10–2016.1 Study the theory of hardware fingerprint and Physical

Unclonable Functions (PUFs).

2016.2–2016.5 Design the new scheme of M2M authentication with the use

of clock skew of wireless network hardware.

2016.06–2016.08 Implement prototypal systems and evaluate the

performance of them.

2016.09–2016.10 Write patents, papers and reports and finish the acceptance

check.

161 / 206


Subject: Key Establishment and Generation with

Wireless Channel CSI in M2M of IoT

M2M System Model of IoT

Securing wireless communication remains challenging in dynamic mobile

environments due to the shared nature of wireless medium and lacking of fixed

key management infrastructures. The underlying channel response between

any two parties of a physical channel is unique and decorrelates rapidly in

space. Generating secret keys using physical layer information thus has drawn

much attention to complement traditional cryptographic-based methods.

Although recent work has demonstrated that Received Signal Strength (RSS)

based secret key extraction is practical, existing RSS-based key generation

techniques are largely limited in the rate they generate secret bits and are

mainly applicable to mobile wireless networks.

In this project, we take a different view point by exploring fine-grained physical

layer information made available from Orthogonal Frequency-Division

Multiplexing (OFDM). The channel response from multiple subcarriers of

OFDM provides detailed Channel State Information (CSI), which can be

162 / 206

utilized to achieve higher secret bit generation rate and make the secret key

extraction approaches (based on physical-layer characteristics) more practical.


M2M: Machine to Machine


RSS: Received Signal Strength

CSI: Channel State Information

MIMO: Multi Input Multi Output

OFDM: Orthogonal Frequency-Division Multiplexing

Background

Exploiting physical layer information for secret key establishment between

wireless devices has attracted much attention recently. The basic idea of

physical layer based secret key extraction is that, through public information

exchange, a pair of wireless devices (e.g., Alice and Bob) can obtain reciprocal

observations on the temporal and spatial randomness of the wireless channel

between them, which are served as the basis for secret key generation.

Different from traditional cryptographic-based methods relying on

computational hardness, the essential security of the secret key generated

from physical layer information of a radio channel is guaranteed by the fact that

the wireless channel between two devices is uncorrelated from other channels.

It thus appears promising that physical layer based secret key generation

methods can be deployed as alternatives of existing encryption methods for

mobile wireless devices with limited resources or without key management

infrastructures (e.g., peer-to-peer association, neighborhood devices changing

frequently).

163 / 206

Existing implementations mainly use Received Signal Strength (RSS) and

channel response extracted from a single frequency to perform key

extraction. And most recent studies focus on improving the secret bit

generation rate by exploiting temporal and spatial variations of radio channel,

multiple antenna diversity, and multiple frequencies). However, since both RSS

and channel response extracted from a single frequency can only provide

coarse-grained information of the radio channel (e.g., each wireless packet

can only provide a single RSS value), the current implementations are largely

limited in their real-world deployment even with the assistance of multi-bit

quantization.

We take a different view point by exploring fine-grained physical layer

information made available from Orthogonal Frequency-Division Multiplexing

(OFDM). The channel response from multiple subcarriers of OFDM provides

detailed Channel State Information (CSI), which can be utilized to achieve

higher secret bit generation rate and make the secret key extraction

approaches (based on physical-layer characteristics) more practical.

We investigate the practical application of utilizing CSI to perform secret key

extraction by exploiting OFDM subcarriers, which could provide fine-grained

channel response information to facilitate significantly higher bit generation

rate as compared with many exiting studies such as the popular RSS based

methods.By leveraging the detailed channel response information, our

proposed approach is resilient to attack scenarios which have been identified

harmful to secret key extraction when using RSS, including predictable

channel attack and stalking attack.

Scope

The randomness of radio channel’s physical layer characteristics have been

theoretically explored for secret key generation. A number of studies have

proposed to use the phase change of received signals to generate secret keys.

164 / 206

Sayeed et. al. and Wilson et. al. exploit the randomness of phase for secret

key extraction in OFDM and UWB systems respectively, whereas Wang et. al.

propose a phase-based scalable and efficient secret key generation scheme.

Tope et. al. utilize the randomness of received signal’s envelope to share the

secrecy between two parties. Similarly, secret bits have been extracted from

the deep fades of channel gain caused by multipath. All of these investigations

are based on theoretical analysis and only provide simulation results.

RSS has been widely used in the work proposing practical secret key

generation methods, because it is readily available in existing wireless

infrastructures. Previous studies on RSS based methods mainly focus on

exploiting temporal and spatial variations of radio channel, and multiple

antenna diversity for secret bit extraction. Since RSS can only provide

coarse-grained channel information, RSS based methods suffer from low

secret bit generation rate. Channel response has also been exploited to

generate secret keys. For example, Mathur et. al. utilize the channel impulse

response (CIR) extracted from a single frequency to generate at most one

secret bit per second. Recent work using frequency selectivity of channel

fading shows the feasibility of generating secret keys in static wireless sensor

networks, however, they only evaluate the entropy of the secret keys and do

not provide the secret bit generation rate in practical environments.

Different from the aforementioned work, we seek to exploit the fine-grained

channel response information provided by OFDM to improve the practical

usage of secret key extraction based on physical layer features. While the

feasibility of using channel state information in OFDM system to generate

secret keys has been explored, these studies have set the theoretic basis and

do not provide any practical solution. Our approach investigates the

non-reciprocities and unknown fading statistics encountered in real

environments. Additionally, our method is resilient to the malicious attacks

which are harmful to RSS based secret key generation.

165 / 206


1 patents:

1) New key generation method with the CSI and error correcting code.

1 prototypal system:

1) Key Establishment and Generation with Wireless Channel CSI in M2M;

1-2 Papers

Acceptance Criteria

Minimal Bit rate of key generation: 64 bits/s

Mismatch rate of key generation: 90%

Minimum number of packets needed to converge to a clock skew: 400.

Phased Project Plan

2015.9–2016.1 Implement existing scheme using RSS and evaluate the

performance;

2016.1–2016.5 Design the new scheme using CSI and prepare the

equipment for performance test and experiments;

2015.6–2016.8 Implement two prototypal systems and evaluate the

performance of them.

2016.08–2016.09 Write patents, papers and reports and finish the

acceptance check.

166 / 206

Engineering Technology（HIRPO201509）


Subject: New Electrolytes and Electrodes’ Surface

Modification in LMB


Lithium Metal Batteries=LMB

Background

High energy density batteries are required as power source for applications

such as telecommunications, portable electronic devices and hybrid electric

vehicles. Lithium ion batteries are currently the batteries of choice to obtain

these objectives. The next generation LMB(with lithium metal anodes rather

than carbon intercalation anodes) are also attractive because of their longer

life, higher energy density and safety. To date, there are quite some challenges

in terms of low energy density, low ionic conductivity, poor cycling performance

and losses of anodes in lithium metal batteries field. To achieve higher energy

density and reliability, electrolyte additives, ionic-liquid electrolytes, polymer

electrolytes and surface modified electrodes of the lithium metal batteries

appear to be the most promising research hotspots.

Scope

The project will aim to achieving higher energy densities and reliability, by

improving ionic conductivity of electrolytes, protecting electrodes, or preparing

new electrolytes. The project should be focused on electrolyte additives,

ionic-liquid electrolytes, polymer electrolytes, or protection and surface

modification of electrodes.

167 / 206


One SCI thesis OR one patent

Research paper about lithium metal batteries and reports on research

progress

One prototype of LMB

Acceptance Criteria

One SCI thesis OR one patent

Research paper about lithium metal batteries: a review of lithium metal

batteries containing principle and development of lithium metal batteries.

Interim reports on research progress containing experiments, achievements

and research plans for next stage

One prototype of LMB：1) the prototype should contain the necessary materials,

such as cathode, anode, and electrolyte; 2)its energy density can reached

international advanced standards. For example, keeping all parameters and

constraints, target values above 500Wh/kg is expected to be met by

non-aqueous cells even if they include an O2 tank, while aqueous cells might

fall short of such requirements, despite the higher operational voltage。

Meanwhile，the prototypes can be charged to 80% charge retention after 100

cycles; 3) the data of the prototype can be sent to our lab and validated by our

lab.

168 / 206


Subject: Research on energy information physical

fusion system

Background

Energy is one of most challenging issues for sustainability of human society in

21 century. Electrical power production mainly based on fossil fuel is soon

coming to an end due to fast growing demand and very low efficiency. In

recent years, wind generation is becoming major energy source to replace

conventional fossil power and reduce greenhouse gas emissions in many

countries. However, how to deal with large amounts of intermittent

generation resources connected to electric power grids have been troubling

power system planners and operators for years. The recent studies indicated

that the reserve requirements and operating cost will increase significantly

from wind variability and uncertainty for a system with high wind penetration.

A good understanding of the stochastic characteristics of wind power is very

important for secured and economic system operation in market environment.

However, the state-of-art technologies including numeric weather prediction

(NWP) based methods, statistical methods, and hybrid approaches can not

give the good enough generation estimation results for system operation.

Cyber-physical energy systems, which can combine cyber space and physical

space of energy system with rich-sensor network, bi-communication

information network and direct-control units, make a good chance to get the

stochastic characteristics of wind power in an innovation way.

This project focus on an advanced research of aggregated generation

estimation of geographically distributed wind farms with cyber-physical energy

systems.

169 / 206

Scope

P (t )u(t ), v(t )

ξ2ξ1

Senor Network

SCADA/EMS

Wind Forecast

Advection System

Thrbulence

Near Surface Wind Field

Wind-power Transition

Σ

Wind Speed Measure

Wind Generation Measure

NWP

Fig. 1 Framework of generation estimation of wind farms

The project including three parts:

i. The system formulation model of aggregated generation estimation of with

cyber-physical energy systems.

ii The advanced research of aggregated generation estimation with given

senor networks

iii A simulation Demo with given area.



ii. A simulation Demo

Acceptance Criteria


ii. A simulation Demo

170 / 206

Phased Project Plan

i. Formulation: 4 months

ii. Simulation with given sensor points: 6 months

iii Research report: 2 months

171 / 206


Subject: Sensor based Innovative interactive

application design for Smart phone


Background

As nowadays smart phone hardware deployment unceasingly rising, the use

experience is also getting better and better. In order to make human-machine

interaction more direct and more convenient, traditional touch screen

interaction are replaced smoothly by some novel mode based on built-in

sensors. But we still need to introduce more new sensors and enrich the user

experience.

Scope



of Sensor based Innovative interactive application design for Smart phone.

2. Implementing a prototype for Sensor based Innovative interactive

application design for Smart phone.


1. One patents;

2. One paper;

3. One technical report with respect to the design and implementation;

4. Prototype for the proposed methods.

Acceptance Criteria

Fail: No patents/papers/prototype are delivered.

172 / 206

Pass:


1 paper is accepted at a IEEE flagship conference/transaction AND

1 detailed technical report AND corresponding prototype

Excellent:


More than 1 paper is accepted by IEEE flagship conferences/transactions.

Phased Project Plan

Phase 1: T+4, deliver a patent and a survey on Sensor based Innovative

interactive application design for Smart phone;

Phase 2: T+8, deliver one technical report;

Phase 3: T+12, deliver one paper, and prototype.

173 / 206


Subject: Exploring Metamaterial to improve antenna

array performance

Background and Motivation

Massive MIMO and beam sweeping will be use to improve communication

capacity, there will be so many challenge in big antenna array design, and

Metamaterial technology maybe a good choice to improve antenna array

performance.

Scope

m*n antenna array, bandwidth >8% ,three problem to be solved:

1) use metamaterial to improve the gain and x-polarization at the +/-60deg

beam sweep in horizontal

2) reduce the ripple wave on the radiation pattern of anyone column antenna

3) to improve isolation between column

Expected Deliverables

Theory analysis report, Simulation and measurement report, Testing module,

Patent

Period

12 months

174 / 206


Subject: Low PIM and High Power Compact wideband

coupler using suspended transmission line


MBB, Big data, Multi-band and multi-mode Antenna deployment,

high-integration, high-capacity network, reduce site space, Feeding network

high integration of Basestation antenna

Scope

Studying small size low PIM and high power capacity 90 and 180 degree

coupler using specific suspended transmission line, 58% bandwidth.

Proposal Suspended line Structure:



Patent

Period

Period:9~12 months

175 / 206


Subject: Low PIM and High Power Compact combiner

using suspended transmission line


MBB, Big data, Multi-band and multi-mode Antenna deployment,

high-integration, high-capacity network, reduce site space, Feeding network

high integration of Basestation antenna

Scope

Microwave compact combiner at 3GHz, low PIM, microstrip at low power, and

specific suspended transmission line at high power

Band1:normalized frequency band（-1，-0.061）,

Band2: normalized frequency band（0.204，1），

Isolation between band1 and band2: >30dB

Fractional bandwidth of band1+band2: 45%

Suspended line Structure:

176 / 206



Patent

Period

Period:9~12 months

177 / 206


Subject: Compact RF filter under moderate power


The RU should be smaller, and the compact RF filter is required.

Scope

High Q filter, more than 1000@3GHz

Small size

Easy mounted on PCB

High selectivity


Practical samples; design and test report; patent;

Period

Develop period for one year

178 / 206


Subject: Novel methods for crosstalk cancellation in

high-speed parallel bus


With increasing data rate, crosstalk has become a more critical factor in the

performance of high-speed links. Unlike the loss and reflection of a high-speed

channel, crosstalk cannot be mitigated by traditional equalization methods.

High-speed memory devices have run into limitations of the timing margins as

the data rates going up. So, how to mitigate the crosstalk of DDR4 and next

generation DDR become very important in order to reduce the risk of design.

Traditionally, we may increase the spacing of transmission line, reduce the

spacing of the reference plane or add ground via at the place where the

reference planes of the signal occur to changing. But the conventional

methods are going to not meeting our needs. We need to find a new method to

solve this problem as soon as possible.

Scope

The aim of this subject is to find the novel methods for solving the crosstalk of

passive channel for DDR4 and next generation DDR. The methods can be

applied to the package or PCB design for crosstalk cancellation or crosstalk

mitigation. And the signal quality of high-speed parallel bus can be improved

through using the novel methods instead of traditional method.


This project will provide one analysis report which contains related theory,

simulation data, measure data, and so on.

Period

Period:12 months

179 / 206


Subject: Novel static and low frequency shielding

technology used on mobile phone

Background and Motivation Hardware system in smart mobile phone become

denser, smaller and thinner day by day, every millimeter is very precious. In

this system, the magnetic component in power supply, NFC antenna, receiver,

camera, speaker, compass and so on, disturbed each other , nowadays, we

put them far from each other as possible we can, or use big and heavy

shielding case to reduce the interference, but the conventional method is going

to not meeting our needs, we need a new method to solve this problem as

soon as possible.

Scope

Novel methods or new material to reduce the static and low frequency

interference between magnetic components, instead of the conventional

methods, meets the mobile phone’s needs of light, small and thin.


Simulation and analysis reports, measure data, component sample, and

related theory introduction.

Period

1 year

180 / 206


Subject: Research in Current-Carrying Capacity and

Reliability for PCB Traces


PCB:Printed Circuit Board


PCB is a popular interconnection carrier among different components in

electronic equipments, when a PCB trace is powered, current flowing through

it generates heat, creating a temperature rise between the trace and the

surrounding environment. Temperature rise of PCB depends upon the trace

length, trace width, copper thickness, and various factors such as PCB

thickness, dielectric material, amount and adjacency of copper in the board,

heat removal system, etc.

Current carrying capacity is the maximum quantity of electrical current a trace

or via can carry before sustaining immediate or progressive decrease, also

described as current rating or ampacity. With the trend of electronic equipment

becomes more and more compact, thermal concentration could happen, it

would cause internal ambient temperature rising in system, but the

environmental factors affect current-carrying capacity significantly. So it is

necessary to study a methodology/guideline to assess the reliability for

proposal design.

Scope

This work should focus on the research in the analytical methodology and how

the PCB trace length, trace width, copper thickness and various factors such

as PCB thickness, dielectric material, amount and adjacency of copper in the

board, heat removal system affect PCB trace and via temperature rise, it is

181 / 206

recommend that equations or graphs or tables which can show us the relations

of PCB parameters VS PCB temperature rise at a given excited current.

Additionally, it should study how to measure the temperature rise precisely for

PCB trace and via, especially for the internal trace of multi-layer PCB.


We expect to get below deliverables through this project:

Provide PCB ampacity technology investigation report, including:

Analytical methodologies and simulation tools which can be used to predict the

temperature rise for PCB trace/via

Top5 research institute/company and their research area and achievements

Provide factors which are qualified to imply PCB reliability performance

Provide an analytical process/method for predicting PCB trace/via temperature

rise at the given excited current and heat removal system

Provide the suitable test methods which can precisely measuring temperature

rise of PCB trace and via

Provide graphs or tables which can show us the relations of PCB parameters

VS PCB temperature rise at a given excited current and heat removal system.

Execute one training at least before ending the project

Notes：

PCB should be multi-layers(≥6 layers), PCB parameters includes trace length,

trace width, copper thickness, PCB thickness, dielectric material, amount and

adjacency of copper in the board

Period

Development period is about one year

182 / 206


Subject: Research in improving the anti-saturation

ability of the active EMI filter


The difficulty of the promotion of power density lies partly on the improvement

of active components' performance and partly on the passive components'.

The development of the semiconductor devices is obviously to all, rapidly,

while the passive components is relatively slowly. Restricted to high frequency

characteristics of magnetic elements, safety regulations of capacitors etc.

Traditional passive EMI filters acquire no breakthrough in recent years. Using

active devices replacing passive devices, active EMI filters can make the

traditional passive EMI filters in progress with the development of

semiconductor devices. Thus, further promotion of power density is promising.

Due to limitation of output voltage and structure of filter circuits, active devices,

such as integrated operational amplifiers, may be saturated, and the lower the

frequency is, the easier the saturation occurs. Once the saturation occurs, the

filter circuit fails, so improving the anti-saturation ability of the active EMI filter

in low frequency guarantees the reliability service of the EMI filter.

Scope

1. Research on the circuit topology of the AC port active EMI filter, the structure

of the noise cancellation circuit and propose effective resolutions.

2. Guarantee the Common-mode differential loss under 40dB at CE frequency

band (150KHZ~30MHZ), and improve anti-saturation capability in low

frequency as far as possibility, DO NOT increase leak current at 50/60HZ

power frequency .

183 / 206

3. If condition 2 is satisfied, and feedback type structure is employed, the EMI

filter should meet stability requirements.

4. If condition 2 and3 is satisfied, further research the technology of active EMI

filter without common mode current detection magnetic components.



Provide active EMI Suppression technology investigation report.

Provide one active EMI filter Engineering prototype.

Provide the Engineering prototype design report and measurement report.

Execute one training at least before ending the project

Period


184 / 206


Subject: Ultra-high frequency Thin Film power

Inductor(TFI) Testing and Modeling Techniques


The integration and miniaturization of inductors has become a major focus of

the power electronics community as the demand for high-performance

low-volume converter has grown. Efficient power conversion circuits most

often rely on magnetic components for energy storage; however inductors and

transformers are generally the largest and most lossy elements in these

systems. Integrating inductor remains a primary challenge in achieving

monolithic solutions for power conversion. Small and efficient magnetic

components, and thereby converters, can increase the penetration of

energy-saving technologies such as voltage regulator by driving down system

costs while increasing performance and efficiency.

There is a important part of research on thin film inductor’s modeling and test

techniques except of design .It is more challenge for us to test and modeling

the inductor’s characteristics because of its small volume, very small amount

of inductance and other non-linear effect.

So, We want to study the inductor’s modeling and simulation analysis

techniques，Including the method of inductance modeling, fitting and analysis,

the fast selection and convergence of the inductance parameters are carry out.

Scope


1) Study on the testing technology of ultra high frequency thin film inductor:

the operation frequency of inductor upto hundreds mega hertz.

185 / 206

2) Study on the method of permeability measurement of ultra high frequency

thin film inductor, and the way of make a measurement fixture

3) Ultra-high frequency film inductor device-level parametric testing Methods:

Including permeability, inductance , DC resistance, AC resistance, and etc

4) Ultra-high frequency inductor saturation characteristics Technology:

Including the principles and methods of test

5) Ultra-high frequency inductance modeling and simulation analysis

techniques: It includes an inductor modeling, fitting and analysis

6) The loss of high-frequency thin-film inductors Simulation & Testing

Technology: Inductance loss testing, magnetic separation of copper loss ,

simulation analysis and comparison of test



1) Provide the report of testing technology of thin film inductor investigate and

survey in trade.

2) Provide Ultra-high frequency thin film inductor simulation model and report

of the method of modeling analytical methodology

3) Provide the report of Ultra-high frequency thin film inductor inductance

modeling and simulation analysis techniques: It includes an inductor

modeling, fitting and analysis

4) Provide the report of Ultra-high frequency thin film inductors Simulation &

Testing Technology: Inductance loss testing, magnetic separation of

copper loss , simulationanalysis and comparison of test.

5) Provide the prototype of test fixture.

6) Execute one training at least before ending the project

186 / 206

Period


187 / 206


Subject: Design,manufacture and performance study

technologies for 20MHz+ high frequency power

supplies and low profile magnetic component


Inductive DC-DC converters currently require the energy storage passives (i.e.

high value input and output capacitors and output inductor) to be discrete,

surface mount, components which have to be assembled off-chip on the

printed circuit motherboard, the external passives take up significant

motherboard footprint and add significantly to the cost of the component bill of

materials. This inability to integrate the power passive components for high

efficiency DC-DC converters is seen as a major roadblock in addressing the

above challenges for future integrated power management systems in

consumer and industrial electronics product design.

Scope

A strategic goal of the project is to develop an end-to-end capabilities (design,

manufacturing, performance study) for 20MHz+ high frequency power

conversion and low profile magnetic component.

A scope of work should include the following components:

1.Research novel circuit-topology for 20MHz+ high frequency power supplies

2.Research and develop materials can have constant permeability at higher

frequencies than conventional materials, such as up to more than 50MHz

3.Provide available technologies for coupled inductor integration

4.Detailed analysis of the performance of magnetic material and coupled

inductor, in terms of coil impedance, core loss and magnetic properties to

providing the best solution

188 / 206


1) Deliver a 20MHz+ high frequency power supply prototype , its typical

power level can be dozens of watts , it can be a isolated or non-isolated

switching power supplies

2) at least two patents

3) deliver a novel circuit-topology for 20MHz+ high frequency power supplies

4) deliver at least one type of high frequency magnetic material has constant

permeability up to more than 50MHz

5) fabricate 20MHz+ high frequency and low loss coupled inductor

6) Very high stability with an outstanding behaviour at 250°C

7) low failure rate, high quality manufacturing process

Period

12 months

189 / 206


Subject: High transparent flexible OLED realization

and innovation application


Flexible OLED display is acknowledged as a next display technology , the high

transparent flexible OLED have a high value using in smart mobile and

wearable device ,but transparency of flexible OLED display only 30% just now

for this technology face some challenge, if transparency exceed 50% ,it will

have a good experience and enhance our competition and difference from

others competitor if the innovation solution realized .

Scope

It should be researched backplane of flexible and transparent pixel pattern

design to achieve the curvature radius, simultaneously to realize performance

requirement as below:

1)The flexible OLED panel has a HD class resolution

2)The panel’s curvature radius is XXR;

3)The transparent OLED panel have an enough transmittance;


High transparent flexible OLED realization and innovation application solution

Period

1year

190 / 206


Subject: The pressure sensing TP


Pressure sensing is one of the cutting edge innovation in the industry.

Pressure sensing can be achieved in many ways, including film, resistors,

sensors, etc. In all these solutions, 2D touching and 3D pressure sensing are

separated.

Scope

Under the normal pressure of finger, such as sensitivity, accuracy, precision,

linearity, reaction speed etc, could fulfilled the need of the latest phone. And in

this solution, 2D and 3D touching are combined.

Pressure sensing specifications are defined by the test on the pressure

sensing TP sample.


the pressure sensing TP

Period

1year

191 / 206


Subject: N in one LGP


LGP:Light Guide Plate


N in one LGP have a great potential for slim smart phone and wearable device

innovation application This BLU has advantages over normal structure BLU;

for example, simple assembly, wrinkle free and no curl. This will be a

revolutionary technology for LCD BLU.

Normal structure: Prism Sheet + Prism sheet + Diffuser + LGP + Reflector

sheet

Two in one: Two in one sheet(3m/LMS) + Diffuser + LGP + Reflector sheet

Three in one: Three in one sheet + LGP + Reflector sheet

N in one LGP: Prism sheet + N in one LGP + Reflector or N in one LGP +

Reflector

Scope

N in one light guide plate realize solution with BLU Thickness: thinner than 50%

thickness of normal structure ,Luminance: the same as flagship phone，

Luminance Uniformity: the same as flagship phone, NTSC: better than

AMOLED.


Slim BLU realize solution / Technical Feasibility Analysis

Patent

Sample test report

192 / 206

Period

1year

193 / 206


Subject: New type IR Camera


With eye tracking a device can understand what the user pays attention to,

what information the person has processed, and it can give indications about

the person’s presence, attention, focus, drowsiness, consciousness or other

mental states. This information can be used to gain deep insights into

consumer behavior or design new user interfaces for games or others

application. Iris recognition is an automated method of biometric identification

that uses mathematical pattern-recognition techniques on video images of one

or both of the irises of an individual's eyes, whose complex random patterns

are unique, stable, and can be seen from some distance. it is possible to

establish an identity based on who you are, rather than by what you possess,

such as an ID card, or what you know, such as a password. Since biometrics

associates the identity of a person with his unique body measurement, it is

much more reliable than non-biometric methods such as password, PIN, and

ID cards.

Scope

Normal and IR photography merge in one camera, high sensitivity without

decrease of resolution, simultaneously increase of sensor size and the field of

view.


IR camera realize solution

Patent

Sample test report

194 / 206

Period

1~2year

195 / 206


Subject: Research on the silicon-based SiGe device

integration


Silicon is indirect band gap material, resulting in the low illumination efficiency.

However, Ge illumination has been achieved by transiting the indirect band to

direct band through some methods, such as high doping. Besides, Ge photo

detector has been commercialized and Ge modulator based on Frank-Keldysh

effect has been reported. Thus it is promising to integrate the light source,

modulator and photo detector with SiGe material.

Scope

1) Study and analysis of the Ge light source, SiGe modulator and Ge Photo

detector;

2) Simulation of Ge light source, SiGe modulator and Ge Photo detector;

3) Fabrication of the integrated chips, including Ge light source, SiGe

modulator and Ge Photo detector.


1) Investigation report of the Ge-based light source, modulator and Photo

detector;

2) Investigation report of the silicon-based integration technology status;

3) Simulation report of Ge light source, SiGe modulator and Ge Photo

detector;

4) Samples of integrated devices including light source, modulator and photo

detector based on SiGe material.

5) A patent as least

196 / 206

Period

1~2 year

197 / 206


Subject: High output power hign operation

temperature laser source


The optical transceiver used in un-control environment (such as wireless

application ) increased rapidly, the operation temperature range from -40℃

to+85℃, some special case more than 95℃. In that case , the laser chip

junction temperature will be higher than 95℃ or upto 100℃ respectively.

Meanwhile，Silicon photonics as a potential lower cost technology for future

optical communication are hot topic in research and investment , and the

silicon photonics also need high power and high temperature operation laser

source because of the big insertion loss within the silicon chip and high power

dissipation.

Scope

1,1.3um band, high output power(up to 20mW ) and high operate temperature

(more than 95℃) un-cool FP/DFB CW laser source for Silicon photonics

application .

2,1.3um band, high speed (up to 25G bps) and high operate temperature(more

than 95℃) un-cool DML (directly modulated laser) for high speed transceiver

in wireless application.


1,Semiconductor laser device meet high output power / high temperature/ high

bit rate requirements list above;

Period

1~2 year

198 / 206


Subject: Analog Laser for ROF with 6~10GHz

bandwidth at 85oC


Analog laser has been widely used in the field of CATV and communication,

but currently analog laser used in ROF only can support no more than 6 GHz

bandwidth and no more than 70 degrees working temperature; With

communication capacity improvement and high temperature requirement in

wireless base station, analog laser need to be developed to a higher

bandwidth and higher working temperature.

Scope

Research scope content: Laser chip development, samples build and test, RF

circuit design;

The problem to be solved: RF performance when bandwidth is more than

6GHz at 85oC


Research report of design, structure and process of the analog laser.

Analog laser chip samples and test report.

At least 1 patentable ideas report.

Period

1 year

199 / 206


Subject: Effect of the metallization on the quality factor

of dielectric resonator


Currently, Dielectric components are more and more popular in RF

applications in terms of size and cost benefits .the main process of

metallization on dielectric ceramic is thick film by silver sintering. But its high

loss blocks many applications with small margin of performance, such as low

cost and miniaturized RF filters. Low loss metallization materials and process

on dielectric materials is more and more important, which can reduce the

power dissipation of the electronic components.

On one hand, the balance of conductivity and peeling strength of the

metallization is one of most challengeable things; higher conductive and strong

metallization is intensely needed in industry. On the other hand, the effect of

interface microstructure on the RF performance is also need to study, because

of skin effect of RF signal.

By Studying metallization materials and process for dielectric ceramic

components, improve the quality factor of the dielectric resonator.

1) Thickness > 10um

2) Electricity Resistance <2.5μ Ω .cm (Currently 5.0 μ Ω .cm）.

3) Peeling strength > 1kgf

4) Solder-able

Scope

1. High conductivity metallization materials and process corresponding.

200 / 206

2. The metallization layer is solder-able and can meet the Soldering

resistance requirement

3. The conductivity of the inner side is more important than the one of the

outside because of skin effect of RF signal. So the balance of conductivity

and peeling strength is the challenge. For example, the glass phase in the

silver thin film layer is one of the key factors.

4. The microstructure of the interface may influence the power dissipation of

the RF signal, like the roughness, the defects, which need to be studied as

well.

5. One kind of resonator samples will be provided for evaluation.


1. Metalized samples that has 40MS( pure silver is around 60)

2. Study report: Effect of the metallization on the quality factor of dielectric

resonator

Period

1.5 years;

201 / 206


Subject: Research on Micro displacement during

Glue/Solder solidification process


Optical module has attracted much attention due to can support optical

network transmission of large information with very high speed. Optical

package includes many kinds of part, such as light source, optical component,

and optical fiber. For solidification process are mainly two solutions, soldered

and Glue for optical module assembly. During assembly, unwanted stress in

the module causes the part to move relative to the optical components,

thereby lowering the coupling efficiency.

Recently optical assembly accuracy requirement is less than 0.5μ m, such as

silicon photonics assembly process. To realize high accuracy solidification

process is the most difficult issue. The objective of this project is to investigate

the theory of micro displacement during solidification process, how to

improvement of this issue for optical module assembly technologies.

Scope

Research content:

1) Analysis of key factor of micro displacement during Glue/Solder

solidification process

2) How to decrease micro displacement in solidification process and future

improvement.

3) How to control micro displacement by Glue process

4) How to control micro displacement by soldered process

Research scope， content，and the problem to be solved

202 / 206


Technical report for experiment design and Simulation, Test Result, Process

condition,

Period

1year

203 / 206


Subject: A kind of component which has positive

current/temperature coefficient characteristic

Background

Serial protective circuit or component (or Semi conduct component circuit) as

same as PTC which can protect sensitive circuit ( IC or power module) ; it

MUST quickly respond for example during lightning surge occur (10us-1ms);

The kind of material or component (or Semi conduct component circuit)

which has high impedance or open circuit when suffers overvoltage or over

current , and MUST be restorable for the circuit ( IC or power module) keep

normal working state later.

When the circuit ( IC or power module) is working in normal condition, the

material or component (or Semi conduct component circuit) has low

impedance, thus No influence on circuit (IC or power module).

Scope

Same as above


A kind of material or component (or Semi conduct component circuit) which

have following specifications:

1. When circuit is working in normal condition（for example,0-20A working

current, -40~+85℃ environment temperature）,the material or component

should have low impedance（for example <20mΩ ）.

2. When the circuit has overvoltage or overcurrent（for example >200A，last

10us-1ms），the material should have quick response（be hot or action）

204 / 206

to realize the rapid increase of impedance (for example >200mΩ , to cut

down and decrease the overcurrent). The material or component is

required to be high impedance within 20us.

3. It should bear high voltage and current; After it respond, it can bear 200V

voltage and surge power(200V*200V/R) which lasts 10us-1ms;

4. When the overcurrent and overvoltage disappeared, it can restore low

impedance quickly（for example in 1ms-1s）;

5. Form: it can be encapsulated like resistance、IC、MOS, etc.

6. Volume：< 10mm*20mm*8mm

Research and test report

Acceptance Criteria

Same as above list 6: Expected Outcome and Deliverables

Phased Project Plan

1~2 years

205 / 206


Subject: Fire disaster protection technology for short

circuit of lithium battery energy storage system


Li-ion ---lithium battery

LiCOO2---Cobalt acid lithium battery

LiFePO4---lithium iron phosphate battery

Background

Lithium battery has obvious advantages compared to conventional cells, for

example high energy density, environmentally friendly. Therefore it is used

more and more widely in modern life, such as terminal equipment,

communication apparatus, electric vehicle, etc.

However, the fire and explosive risk of lithium battery caused by short circuit

cannot be ignored. Numerous accident cases have aroused the attention of

the public and people of this field. Therefore，investigation on the protection

technology for short circuit of lithium battery energy storage system is

necessary and meaningful.

Solution to prevent fire caused by short circuit and protection technology for

lithium battery is in search.

Scope

Research scope (Including but not limited to):

Single cell of lithium battery or lithium battery pack (7.2Wh~100kWh)

The materials of lithium battery are not limited, such as cobalt acid and lithium

iron phosphate, etc.

206 / 206


Looking forward to acquire the most frontier protection technology for

preventing fire caused by the short circuit of lithium battery. Including but not

limited to:

1. Safety material of single cell of lithium battery, flame resistant material of

battery positive electrode, separator material excellent at thermal

shutdown performance

2. The best series-parallel connection methods and theory of the single cells

in a energy storage system

3. The protection methods and electric circuit of the anti short circuit for the

lithium battery energy storage system

4. The solution for the short circuit caused fire of the lithium battery energy

storage system used in outdoor

Research and test report

Acceptance Criteria

Same as above list 6: Expected Outcome and Deliverables

Phased Project Plan

1~2 years