bats: network coding in action - istc2018.org · bats: network coding in action raymond w. yeung...
TRANSCRIPT
BATS: Network Coding in Action
Raymond W. Yeung
Institute of Network CodingThe Chinese University of Hong Kong
December 1, 2018
Joint work with Shenghao Yang, Hoover Yin, Tsz-Ching Ng, et al.
R. W. Yeung (CUHK) BATS December 1, 2018 1 / 31
Outline
1 File Transmission in Networks with Packet Loss
2 Random Linear Network Coding
3 BATS Codes
4 BATS Protocol and Implementations
5 Smart Cities Applications
R. W. Yeung (CUHK) BATS December 1, 2018 2 / 31
Networks with Packet Loss (Erasure Networks)
One 20MB file ⇡ 20,000 packets
A practical solution
low computational and storagecosts
high transmission rate
small protocol overhead
x1 x2 · · · xK
s
t1 t2
R. W. Yeung (CUHK) BATS December 1, 2018 3 / 31
Networks with Packet Loss (Erasure Networks)
One 20MB file ⇡ 20,000 packets
A practical solution
low computational and storagecosts
high transmission rate
small protocol overhead
x1 x2 · · · xK
s
t1 t2
R. W. Yeung (CUHK) BATS December 1, 2018 3 / 31
Smart City
http://www.monitormymeter.com/lighting-automation.html
R. W. Yeung (CUHK) BATS December 1, 2018 4 / 31
Vehicular ad-hoc Network
R. W. Yeung (CUHK) BATS December 1, 2018 5 / 31
Balloon-powered Internet
http://www.theneweconomy.com/insight/googles-project-loon-explained
R. W. Yeung (CUHK) BATS December 1, 2018 6 / 31
Low-orbit Satellite Internet
http://www.cs.ucsb.edu/~almeroth/classes/S99.290I/Satellite.html
R. W. Yeung (CUHK) BATS December 1, 2018 7 / 31
Underwater Acoustic Network
R. W. Yeung (CUHK) BATS December 1, 2018 8 / 31
Power-line Communication Network
R. W. Yeung (CUHK) BATS December 1, 2018 9 / 31
Why Loss and Relay are Inevitable
WLAN has more and more interference in both 2.4GHz and 5GHzHigher loss due to interference
Higher frequency in millimeter wave spectrum e.g. 60 GHz to beadopted in 802.11AD and 5G
Higher loss due to obstaclesRelay for non-line-of-sight transmission
Low power transmission in IoTHigher loss due to low S/N ratioRelay for long distance transmission
Underwater acoustic communicationHigh loss due to path loss, noise, multi-path, Doppler spreadRelay for long distance transmission
R. W. Yeung (CUHK) BATS December 1, 2018 10 / 31
Line Networks of n Hops
s r1 r2 · · ·
rn�1
t
All links have a packet loss rate 0.2.
Intermediate Operation Maximum Rate
forwarding 0.8n ! 0network coding 0.8
In practice, it is very di�cult to build a line network with morethan 5 or 6 hops – the “multi-hop curse”
R. W. Yeung (CUHK) BATS December 1, 2018 11 / 31
Line Networks of n Hops
s r1 r2 · · ·
rn�1
t
All links have a packet loss rate 0.2.
Intermediate Operation Maximum Rate
forwarding 0.8n ! 0network coding 0.8
In practice, it is very di�cult to build a line network with morethan 5 or 6 hops – the “multi-hop curse”
R. W. Yeung (CUHK) BATS December 1, 2018 11 / 31
Line Networks of n Hops
s r1 r2 · · ·
rn�1
t
All links have a packet loss rate 0.2.
Intermediate Operation Maximum Rate
forwarding 0.8n ! 0network coding 0.8
In practice, it is very di�cult to build a line network with morethan 5 or 6 hops – the “multi-hop curse”
R. W. Yeung (CUHK) BATS December 1, 2018 11 / 31
Network Coding is a Necessity
Link-by-link retransmissionCache O(
pK) packets at intermediate nodes.
Feedbacks may not be reliable or available.Far from optimal for multicast.
Link-by-link erasure coding (fountain coding)Cache O(K) packets at intermediate nodes.High computation costs at intermediate nodes and/or destinationnodes.Far from optimal for multicast.
R. W. Yeung (CUHK) BATS December 1, 2018 12 / 31
Outline
1 File Transmission in Networks with Packet Loss
2 Random Linear Network Coding
3 BATS Codes
4 BATS Protocol and Implementations
5 Smart Cities Applications
R. W. Yeung (CUHK) BATS December 1, 2018 13 / 31
Multicast Capacity of Erasure Networks
Random linear network codes achieve the capacity of a large range ofmulticast erasure networks.
However, the complexity issues prevent the real-world implementationof the baseline RLNC scheme.
[Wu06] Y. Wu, “A trellis connectivity analysis of random linear network coding with bu↵ering,” in Proc. IEEE ISIT 06, Seattle,USA, Jul. 2006.
LMKE08] D. S. Lun, M. Medard, R. Koetter, and M. E↵ros, “On coding for reliable communication over packet networks,”Physical Communication, vol. 1, no. 1, pp. 320, 2008.
R. W. Yeung (CUHK) BATS December 1, 2018 14 / 31
Multicast Capacity of Erasure Networks
Random linear network codes achieve the capacity of a large range ofmulticast erasure networks.
However, the complexity issues prevent the real-world implementationof the baseline RLNC scheme.
[Wu06] Y. Wu, “A trellis connectivity analysis of random linear network coding with bu↵ering,” in Proc. IEEE ISIT 06, Seattle,USA, Jul. 2006.
LMKE08] D. S. Lun, M. Medard, R. Koetter, and M. E↵ros, “On coding for reliable communication over packet networks,”Physical Communication, vol. 1, no. 1, pp. 320, 2008.
R. W. Yeung (CUHK) BATS December 1, 2018 14 / 31
Complexity of Linear Network Coding
CV overhead: K/T .
Encoding: O(TK) per packet.
Decoding: O(K2 + TK) per packet.
Network coding: O(TK) per packet. Bu↵er K packets.
encoding
network coding
K: number of packets. T : packet length.R. W. Yeung (CUHK) BATS December 1, 2018 15 / 31
Outline
1 File Transmission in Networks with Packet Loss
2 Random Linear Network Coding
3 BATS Codes
4 BATS Protocol and Implementations
5 Smart Cities Applications
R. W. Yeung (CUHK) BATS December 1, 2018 16 / 31
BATched Sparse (BATS) Codes
outer code
inner code(network code)
[YY11] S. Yang and R. W. Yeung. Coding for a network coded fountain. ISIT 2011, Saint Petersburg, Russia, 2011.
[YY14] S. Yang and R. W. Yeung. Batched sparse codes. Information Theory, IEEE Transactions on, vol. 60, no. 9, pp.5322-5346, Sep. 2014.
[YY17] S. Yang and R. W. Yeung. BATS Codes: Theory and Practice, in Synthesis Lectures on Communication Networks,Series Editor: R. Srikant. Morgan & Claypool Publishers, 2017.
R. W. Yeung (CUHK) BATS December 1, 2018 17 / 31
BATS Codes Theory and Practice
Shenghao YangRaymond W. Yeung
Series ISSN: 1939-4608
Synthesis Lectures onCommunication Networks
Series Editor: R. Srikant, University of Illinois at Urbana-Champaign
Analytical Methods for Network Congestion ControlShenghao Yang, The Chinese University of Hong Kong, ShenzhenRaymond W. Yeung, The Chinese University of Hong KongThis book discusses an efficient random linear network coding scheme, called BATched Sparse code, or BATS code, which is proposed for communication through multi-hop networks with packet loss. Multi-hop wireless networks have applications in the Internet of Things (IoT), space, and underwater network communications, where the packet loss rate per network link is high, and feedbacks have long delays and are unreliable. Traditional schemes like retransmission and fountain codes are not sufficient to resolve the packet loss so that the existing communication solutions for multi-hop wireless networks have either long delay or low throughput when the network length is longer than a few hops. These issues can be resolved by employing network coding in the network, but the high computational and storage costs of such schemes prohibit their implementation in many devices, in particular, IoT devices that typically have low computational power and very limited storage. A BATS code consists of an outer code and an inner code. As a matrix generalization of a fountain code, the outer code generates a potentially unlimited number of batches, each of which consists of a certain number (called the batch size) of coded packets. The inner code comprises (random) linear network coding at the intermediate network nodes, which is applied on packets belonging to the same batch. When the batch size is 1, the outer code reduces to an LT code (or Raptor code if precode is applied), and network coding of the batches reduces to packet forwarding. BATS codes preserve the salient features of fountain codes, in particular, their rateless property and low encoding/decoding complexity. BATS codes also achieve the throughput gain of random linear network coding. This book focuses on the fundamental features and performance analysis of BATS codes, and includes some guidelines and examples on how to design a network protocol using BATS codes.
Synthesis Lectures onCommunication Networks
store.morganclaypool.comR. Srikant, Series Editor
YANG • YEUNG
BATS CODES: THEORY AND PRACTICE
M
OR
GA
N &
CLAY
POO
L
About SYNTHESIS
This volume is a printed version of a work that appears in the Synthesis Digital Library of Engineering and Computer Science. Synthesis books provide concise, original presentations of important research and development topics, published quickly, in digital and print formats.
R. W. Yeung (CUHK) BATS December 1, 2018 18 / 31
Encoding of BATS Code: Outer Code
Apply a “matrix fountain code” at the source node:1 Obtain a degree d by sampling a degree distribution .2 Pick d distinct input packets randomly.3 Generate a batch of M coded packets using the d packets.
Transmit the batches sequentially.
b1 b2 b3 b4 b5 b6
· · · · · ·
X1 X2 X3 X4
Xi =⇥bi1 bi2 · · · bidi
⇤Gi = BiGi.
R. W. Yeung (CUHK) BATS December 1, 2018 19 / 31
Encoding of BATS Code: Inner Code
The batches traverse the network.
Encoding at the intermediate nodes forms the inner code.
Linear network coding is applied in a causal manner within a batch.
snetwork with linearnetwork coding
t
· · · , X3, X2, X1 · · · , Y3, Y2, Y1
Yi = XiHi, i = 1, 2, . . ..
R. W. Yeung (CUHK) BATS December 1, 2018 20 / 31
Belief Propagation Decoding
1 Find a check node i with degreei = rank(GiHi).
2 Decode the ith batch.
3 Update the decoding graph. Repeat 1).
b1 b2 b3 b4 b5 b6
G1H1 G2H2 G3H3 G4H4 G5H5
The linear equation associated with a check node: Yi = BiGiHi.
R. W. Yeung (CUHK) BATS December 1, 2018 21 / 31
Precoding
Precoding by a fixed-rate erasure correction code.
The BATS code recovers (1� ⌘) of its input packets.
Precode
BATS code
[Shokr06] A. Shokrollahi, Raptor codes, IEEE Trans. Inform. Theory, vol. 52, no. 6, pp. 25512567, Jun. 2006.
R. W. Yeung (CUHK) BATS December 1, 2018 22 / 31
Complexity
Coe�cient vector overhead M/TSource node encoding O(MT ) per packet
Destination node decoding O(M2 +MT ) per packet
Intermediate Nodebu↵er O(MT )
network coding O(MT ) per packet
T : length of a packet
K: number of packets
M : batch size, M ⌧ K,T .
R. W. Yeung (CUHK) BATS December 1, 2018 23 / 31
Achievable Rates for Line Networks: Up to 50 Hops
0 10 20 30 40 500
0.2
0.4
0.6
0.8
network length (l)
R⇤forsystem
atic
recoding
qm = 28, M = 64
qm = 28, M = 32
qm = 28, M = 16
qm = 28, M = 8
qm = 28, M = 4
qm = 28, M = 2
qm = 28, M = 1
R. W. Yeung (CUHK) BATS December 1, 2018 24 / 31
Achievable Rates for Line Networks: Up to 50 Hops
0 10 20 30 40 500
0.2
0.4
0.6
0.8
network length (l)
R⇤forsystem
atic
recoding
qm = 28, M = 64
qm = 28, M = 32
qm = 28, M = 16
qm = 28, M = 8
qm = 28, M = 4
qm = 28, M = 2
qm = 28, M = 1
qm = 2, M = 64
qm = 2, M = 32
qm = 2, M = 16
qm = 2, M = 8
qm = 2, M = 4
qm = 2, M = 2
qm = 2, M = 1
R. W. Yeung (CUHK) BATS December 1, 2018 24 / 31
Achievable Rates for Line Networks: Up to 1000 Hops
0 200 400 600 800 1,0000
0.2
0.4
0.6
0.8
network length (l)
R⇤forsystem
atic
recoding
qm = 28, M = 64
qm = 28, M = 32
qm = 28, M = 16
qm = 28, M = 8
qm = 28, M = 4
qm = 28, M = 2
qm = 28, M = 1
R. W. Yeung (CUHK) BATS December 1, 2018 25 / 31
Achievable Rates for Line Networks: Up to 1000 Hops
0 200 400 600 800 1,0000
0.2
0.4
0.6
0.8
network length (l)
R⇤forsystem
atic
recoding
qm = 28, M = 64
qm = 28, M = 32
qm = 28, M = 16
qm = 28, M = 8
qm = 28, M = 4
qm = 28, M = 2
qm = 28, M = 1
qm = 2, M = 64
qm = 2, M = 32
qm = 2, M = 16
qm = 2, M = 8
qm = 2, M = 4
qm = 2, M = 2
qm = 2, M = 1
R. W. Yeung (CUHK) BATS December 1, 2018 26 / 31
Outline
1 File Transmission in Networks with Packet Loss
2 Random Linear Network Coding
3 BATS Codes
4 BATS Protocol and Implementations
5 Smart Cities Applications
R. W. Yeung (CUHK) BATS December 1, 2018 27 / 31
A Demo for Video Transmission
Video streaming between 2 PC’s through 10 Raspberry Pi 3
11 wireless hops with significant packet loss due to interference
R. W. Yeung (CUHK) BATS December 1, 2018 28 / 31
Outline
1 File Transmission in Networks with Packet Loss
2 Random Linear Network Coding
3 BATS Codes
4 BATS Protocol and Implementations
5 Smart Cities Applications
R. W. Yeung (CUHK) BATS December 1, 2018 29 / 31
Smart Lampposts• Key infrastructure of smart cities• Equipped with networking interfaces, cameras and
sensors• Promote smart city innovations on a city scale– intelligent transportation– autonomous driving– real-time surveillance– high-speed WiFi coverage
• Estimated over 70 million smart lampposts will be installed worldwide by 2027
• Creating a global market of USD 8.3 billion
2
Smart Lamppost Connectivity
• Smart lampposts must be connected to the Internet backbone
• Possible technologies– optical fiber– 4G– BATS
3
Optical Fiber• Pros– very high data rate– highly reliable
• Cons– high installation cost– very long setup time– very disrupting process– sometimes not possible
• Realistically only a small number of lampposts can be connected by optical fiber
• The rest still need to be connected to the Internet
4
How about 4G?
• A 4G card is installed at each lamppost• Pros– easy to deploy– relatively inexpensive
• Cons– high recurrent cost– bandwidth drops drastically during rush hours
5
opticalfiber
The Multi-hop Solution
Advantages of BATS
high throughput
low latency
low coding complexity
low storage requirement
þ
þ
þ
þ
10
Comparison with 4G
12
Low installation cost
Easy to deploy
Low recurrent cost
Guaranteed bandwidth
Reach rural areas
4G ✓ ✓BATS ✓ ✓ ✓ ✓ ✓
Summary
BATS breaks the multi-hop curse in wireless networks
Essentially converts a multi-hop wireless network into a single-hopwireless network
An enabling communication technology for wireless mesh networks,V2X, IoT, smart cities, and beyond
R. W. Yeung (CUHK) BATS December 1, 2018 30 / 31
Reference
[ACLY00] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung. Network information flow. IEEE Trans. Inform. Theory,46(4):1204-1216, July 2000.
[CHKS09] M.-L. Champel, K. Huguenin, A.-M. Kermarrec, and N. L. Scouarnec. LT network codes. Research Report RR-7035,INRIA, 2009.
[CWJ03] P. A. Chou, Y. Wu, and K. Jain. Practical network coding. In Proc. Allerton Conf. Comm., Control, and Computing,Oct. 2003.
[Dana06] A. F. Dana, R. Gowaikar, R. Palanki, B. Hassibi, and M. E↵ros, Capacity of wireless erasure networks, IEEE Trans.Inform. Theory, vol. 52, no. 3, pp. 789804, 2006.
[GS08] R. Gummadi and R. Sreenivas. Relaying a fountain code across multiple nodes. In Proc. IEEE ITW 08, May 2008.
[HB10] A. Heidarzadeh and A. H. Banihashemi. Overlapped chunked network coding. In Proc. ITW 10, pages 1–5, 2010.
[Ho03] T. Ho, B. Leong, M. Medard, R. Koetter, Y. Chang, and M. E↵ros, The benefits of coding over routing in a randomizedsetting, in Proc. IEEE ISIT 03, Jun. 2003.
[LMKE08] D. S. Lun, M. Medard, R. Koetter, and M. E↵ros, “On coding for reliable communication over packet networks,”Physical Communication, vol. 1, no. 1, pp. 320, 2008.
LSS11] Y. Li, E. Soljanin, and P. Spasojevic, E↵ects of the generation size and overlap on throughput and complexity inrandomized linear network coding, IEEE Trans. Inform. Theory, vol. 57, no. 2, pp. 1111–1123, Feb. 2011.
[LYC03] S.-Y. R. Li, R. W. Yeung, and N. Cai. Linear network coding. IEEE Trans. Inform. Theory, 49(2):371381, February2003.
[MAB12] K. Mahdaviani, M. Ardakani, H. Bagheri and C. Tellambura, “Gamma codes: a low-overhead linear-complexity networkcoding solution,” NetCod 2012.
[MHL06] P. Maymounkov, N. J. A. Harvey, and D. S. Lun. Methods for e�cient network coding. In Proc. Allerton Conf. Comm.,Control, and Computing, Sept. 2006.
[PFS05] P. Pakzad, C. Fragouli, and A. Shokrollahi. Coding schemes for line networks. In Proc. IEEE ISIT 05, 2005.
[SZK09] D. Silva, W. Zeng, and F. R. Kschischang. Sparse network coding with overlapping classes. In Proc. NetCod 09, pages74–79, 2009.
[TF11] N. Thomos and P. Frossard, Degree distribution optimization in Raptor network coding, in Proc. IEEE ISIT 11, Aug.2011.
[Wu06] Y. Wu, “A trellis connectivity analysis of random linear network coding with bu↵ering,” in Proc. IEEE ISIT 06, Seattle,USA, Jul. 2006.
R. W. Yeung (CUHK) BATS December 1, 2018 31 / 31