HUAWEI TECHNOLOGIES CO., LTD.

Standardization of Energy Efficiency Measurement

Methods for Network Functions Virtualization

Zhu Bin, bin.zhubin@huawei.com

Huawei Technologies Co. Ltd.

Jun 2015

Page 2 Huawei Confidential

Contents

• Background, Challenges and Opportunities

• Current energy efficiency measurement methods

• Topics need to be studied

Page 3 Huawei Confidential

NFV make telecom network transform from hardware-

centric to software-driven architecture • Hardware and software decoupled, virtualizing network function move

from dedicated network hardware to industry standard server

Virtualized Network Functions (VNFs)

NFV Infrastructure (NFVI)

Physical Infrastructure

Virtual Infrastructure

Compute Storage Network

Virtual Computing Virtual Storage Virtual Networking

NF

V M

an

ag

em

en

t an

d

Orc

hestra

tion

(MA

NO

)

vIMS vEPC vBRAS vRGW/

vAR/vSTB ……

Page 4 Huawei Confidential

NFV Energy Efficiency Requirements

• Workload consolidation

• Resource scheduling

• Energy saving mgmt

mechanism

ETSI GS NFV 004 V1.1.1(2013-10) NFV: Virtualization Requirements

Is it feasible for 50%

energy saving?

Page 5 Huawei Confidential

NFV energy saving challenge and opportunities

Energy waste: Lower

performance / hardware efficiency

Energy saving: Higher

resource utilization

General purpose HW vs.

specified HW

Extra energy consumption

from virtualization layer

Resource sharing, pooling

and provision on demand

Energy aware resource

scheduling to consolidate the

physical machines

Hardly energy efficient co-design

for multi-vendor components in

layered architecture

Page 6 Huawei Confidential

Energy consumption of NFV is dominated by the degree

of resource sharing

Traditional HGW Virtual HGW

No. of HGW 1000 1000

HGW box 18W 6W

Server 0 150W

Total power 18000W 6150W Virtual HGW

- routing and NAT

- firewalling

- OAM management

Quasi-passive CPE

-switch

-Modem/ONT Source: GreenTouch, Virtualizing Home Gateways for large

scale energy reduction in networks, 2012

GreenTouch Demo: Dell server R610 (intel

Xeon5560) hosting with 1000vHGW

- 300% energy saving

Energy saving benefit from:

Time multiplexing of shared processor on server

Elastic resource provisioning on demand

Challenge:

Energy saving gains diminished as energy efficiency of traditional HGW continuous improvement

(15W/2011, 11W/2013, 10W/2015)

Page 7 Huawei Confidential

Limited energy efficiency improvement in virtual IMS

2M subscribers for IMS control plan

NEs (OMU, SPG, CCF, CSCF, ATS)

Core Site

ATS

CSCF

A-SBC I-SBC

OMU/EMS SPG CCF

Node level: energy efficiency improved (35%)

Network level: energy efficiency improved (3%)

HW efficiency improvement (2x)

by new generation processor

performance degradation (-

30%) by VM

overhead for virtualization and

cloud OS

Common infrastructure, e.g.

Fan, Fabric, storage, network

Worse for small capacity Challenge:

virtualization performance improvement

Linear resource scalability

0

0.2

0.4

0.6

0.8

1

1.2

Legacy IMS Cloud IMS

0

0.2

0.4

0.6

0.8

1

1.2

Legacy IMS Cloud IMS

-35%

-3%

Page 8 Huawei Confidential

Energy efficiency evaluation and measurement

• Why? If you can’t measure, you can’t manage and improve

• Procurement: Comparing energy efficiency of different implementation of same

functions (VNF, VNFI, HW)

• Planning: energy efficiency analysis for network deployment and optimization

• Operation: EE visualization for effective energy management

• How? Metrics and Measurement

• Extend EE/EC metrics and measurement methods of telecom equipment to VNF app

• Adapt EE/EC metrics and measurement of cloud/IT system to NFV scenarios

Page 9 Huawei Confidential

Current EE metrics & measurement methods in CT

• Core NW equipment: ETSI ES 201554 (2014-07)

• Mobile core functions (GGSN, HLR, MGW, MME, MSC, SGSN and

PGW/SGW)

• Energy Efficiency metrics is defined as:

• Useful output is maximum capacity of SUT depending on different

functions for voice, data or subscriber

• Average Power Consumption is weighted power consumption for

SUT under different load levels.

TS: (maximum capacity)

• High: TH = 1,0 × TS

• Mid: TM = 0,7 × TS

• Low : TL = 0,1 × TS

Only for All-in-One

equipment

Page 10 Huawei Confidential

Current EE metrics & measurement methods for server

• SPEC benchmarks

• SPECpower_ssj2008: evaluate and comparing server energy efficiency

• Server Energy Efficiency Rating Tool (SERT): ENERGY STAR program

• SPECvirt_sc2013: measure performance of virtualized platforms

Test configuration SERT worklets

not include network-

intensive workloads

15x energy efficiency gain since release

Page 11 Huawei Confidential

Software application energy efficiency comparison

• Microsoft Internet Explorer is most energy efficient browser, consuming 18% less

energy than Google Chrome and Mozilla Firefox

Source: Fraunhofer, the impact of internet browsers on computer energy consumption, 2013

Page 12 Huawei Confidential

Energy Efficiency Metrics

SUT Metrics Definition Measurement method

All-in-one

System EE

System power consumption / system capacity

of typical or max configuration

(e.g. W/ subscribers | PDP | SAU)

Similar with ETSI core NW equipment EE

measurement standard, update needed

VNF

VNF EE

Requirement for vCPU processing capability /

VNF capacity

(e.g. GHz/subscribers | PDP | SAU)

Requirement for vCPU processing

capability = (Number * Frequency) of

cores

VNF

proportional

scalability

VNF EE at maximum capacity / VNF EE at

minimum capacity

Virtualization

layer

Virtualization

EE

Virtualization loss, resource overhead of

hypervisor

Specvirt_ssj2013

Server Server EE

General server EE benchmark, customized

workloads (network intensive) for NFV use

case of data plane

Specpower_ssj2008, SERT

ETSI GS NFV-PER001, NFV

performance & portability best practices

Page 13 Huawei Confidential

VNF EE measurement example

Subscribers 200k 1M 2M 3M 5M

VM numbers 20 60 80 110 150

vCPU Core(2.4GHz) 50 140 180 240 320

VNF EE (GHz/10k

subscribers) 6.0 3.4 2.2 1.9 1.5

VNF proportional scalability 0.25

Resource efficiency of VNF deployment

on specific capacity

Linearity of resource demand on

different VNF capacity

VNF scaling

6.0

3.4

2.2 1.9 1.5

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

200k 1M 2M 3M 5M

Page 14 Huawei Confidential

Summary

• Energy efficiency is still a key issue for NFV transformation

• New energy efficiency measurement methods need to be defined for

layered components in NFV architecture

THANK YOU www.huawei.com

Copyright© 2014 Huawei Technologies Co., Ltd. All Rights Reserved.

The information in this document may contain predictive statements including, without limitation, statements regarding the future

financial and operating results, future product portfolio, new technology, etc. There are a number of factors that could cause

actual results and developments to differ materially from those expressed or implied in the predictive statements. Therefore, such

information is provided for reference purpose only and constitutes neither an offer nor an acceptance. Huawei may change the

information at any time without notice.