network coding for mobile devices and networks: a bme/aau/szte success story frank h.p. fitzek...

Network Coding for Mobile Devices and Networks: A BME/AAU/SZTE Success Story

Frank H.P. FitzekAalborg University

User Cooperation in Mobile Cloud

Reliable Multicast (1)

Reliable Multicast (2)

?

Collector‘s Problem

P = N * ( ln (N) + 0.577)

P

N

sharing

buying

Resource Sharing for Mobile Clouds

010101001010101011101111101011010011011101001

App

Resource Sharing Display (1)

Resource Sharing Display (2)

Taken from the book „Mobile Clouds“, Fitzek/Katz

Resource Sharing Microphone

Single microphone ... ... vs. microphone array

Huge interest by the „Evil“...

Microsoft IPRs 2008 Apple IPRs 2008

… but we cover you …

F.H.P. Fitzek and M. Katz. Cooperation in Wireless Networks: Principles and Applications - Real Egoistic Behavior is to Cooperate!. 2006. Springer.

F.H.P. Fitzek and M. Katz. Cognitive Wireless Networks: Concepts, Methodologies and Visions Inspiring the Age of Enlightenment of Wireless Communications. 2007. Springer.

SRC

DST

R1

R2

R3

R4

50%

50%

50%

50%

0%

0%

0%

0%

Inter flow network coding

XOR type of codingCOPE

CATWOMAN...

Intra flow network coding

RLNC type of codingMORE

...

What is network coding?

b

b

a

a

a b

a+b

a+b

a+b

a a bb

Network Coding: The Butterfly

• Two packets a and b should be conveyed to two destinations

• Capacity per link can handle one packet per time slot

• Bottleneck in the middle

• Either packet a or b will path the bottleneck

b

b

a

a

a b

a a b

a

a

b

b

Network Coding: The Butterfly

• Let‘s try b instead of a• Same old problem

b

b

a

a

a b

a bb

Network Coding: The Butterfly

• Ahlswede et. al. In 2000• Coding the packet• Other ideas were

around• Max-flow min-cut

theorem

b

b

a

a

a b

a+b

a+ba+b

a a bb

Ahlswede, Rudolf; N. Cai, Shuo-Yen Robert Li, and Raymond Wai-Ho Yeung (2000). "Network Information Flow". IEEE Transactions on Information Theory, IT-46 46 (4): 1204–1216.

Network Coding: The ButterflyXOR operation• Bitwise operation• Same bit value results

in „0“• Different bit value

results in „1“

0011

0011

0101

0101

0101 0011

0110

01100110

0101 0101 00110011

CODING

0101 XOR 0011

0110

DECODING

0101 XOR 0110

0011

DECODING

0011 XOR 0110

0101

Network Coding: The Butterfly

Source: transmitting two

information entities

Receiver: receiving two

coded information

entities

Network Coding: The Butterfly

Network Coding: The Butterfly

Network Coding: The Butterfly++

a

a

a

c

a

result = (Rank(A)==3) ? "decodable" : "better choice next time”;

bc

c

c

c

b

ca

a+b

b+c

a+c

a+b+c ?? ?

a

b

b

Network Coding: The Butterfly++

a

a

a

c

a bc

c

c

c

b

ca

a+b

b+c

a+c

a+b+c ?? ?

a

b

b

What did you send? What did you send?

What should we send?

What did you send?What did you send?

What did you recveive?What did you recveive?

Kirchhoff versus Network Coding

All engineers follow this principle!

Now we are alone ...!

store and

forward

computeand

forward

Application fieldsCommunication SystemsP2P (fixed)

Satellites

Reliable Multicast

Channel Bundling

Wireless Mesh

Mobile Mesh

Storage

Distributed Cloud Storage

Random Linear Network Coding

S

R R

D

Psucc = 0.5

a,b a,b

a a,b

No coding

S

R R

D

Psucc = 0.6667

a,b a,b

a a,b,a+b

Binary coding

S

R R

D

Psucc = 1

a,b a,b

αa+ βb α’a+ β‘b

RLNC coding

S60 Implementation RLNC (2007)pre-allocated the memory, generated the encoding vectors, so that we only had the raw encoding

16

32

64

128

1

10

100

1000

10000

2

8

16

OPF

2816OPF

16

32

64

128

1

10

100

1000

10000

2

8

16

OPF

2816OPF

Current coding speeds in 2012

Generation size

Field size

Speed[kByte/s]

... and in 2007 we had 20 kByte/s for generation size of 50!

16

32

64

128

1

10

100

1000

10000

2

8

16

OPF

2816OPF

Current coding speeds in 2012

Generation size

Field size

Speed[kByte/s]

... and in 2007 we had 20 kByte/s for generation size of 50!

First implementations

• P. Vingelmann, P. Zanaty, F.H.P. Fitzek, and H. Charaf, “Implementation of random linear network coding on opengl-enabled graphics cards,” in European Wireless 2009, Aalborg, Denmark, May 2009.

Energy consumption: KODO

Energy consumption: KODO

What about the dynamic changes?

Source

Measurement Setup

Testbed

Inter flow network coding

XOR type of codingCOPE

CATWOMAN...

Intra flow network coding

RLNC type of codingMORE

...

What is network coding?

Inter & Intra flow network coding

Combine the best ...CORE

...

CORE: How does it work?

Identify coding zones/regions

CORE: How does it work?

Identify coding zones/regions

S1

S2

R

D2

D1

Challenge: channel lossesOverhearing is critical for inter-session

Idea: bridge inter- and intra-session network codingAre we the first to think of combining the two?No, it is a natural ideaExample: I2NC by Seferoglu et alBut both coding mechanism are treated in different

layers and are agnostic to each other.• Intra-session: between TCP and IP • Inter-session: above MACWhat makes us different?Code structure provided by RLNC (intra-session) used to

strengthen inter-session

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

CORE

Distinctive Features• Feedback from R to S1 and S2

– Reduces transmissions– May reduce collisions

• Relay can XOR:– New packets from both flows– New packet from a flow with recoded packet from the other

flow (using packets stored @Relay)– Recoded packets from both flows

• Receivers (D1 and D2) can:– Use XOR if overhearing was successful to recover their

intended coded packet– Store XORed and overheard packets– Perform partial decoding for recovering intended coded

packets

Throughput vs Offered Load

Symmetric lossesLoss prob = 0

Throughput vs Offered Load

Symmetric lossesLoss prob = 0.3

Throughput vs Offered Load

Symmetric lossesLoss prob = 0.5

Symmetric lossesLoss prob = 0.8

Throughput vs Offered Load

Demonstrator

• On youtube: http://www.youtube.com/watch?v=mKiHKtZRFVU• (Search for “CORE Network Coding”)

Cooperation Focus

• Network coding sample application filed: FuturICT.hu (BME/AAU/SZTE)

• Related topics:– Mobile cross-platform design patterns– Mobile P2P middleware design patterns

and algorithms– Mobile cloud – Monitoring mobile and sensor networks– Model-driven application development

for multiple mobile platforms

Cooperation Outcome(without having 1 Euro received)

• Combined AAU/BME PhD program– 1 completed– 4 ongoing

• Three book projects and one project with Wiley ongoing• Best Paper award at CCNC 2011 on network coding with iOS• StartUps in Berlin and Aalborg• Combined PhD courses (with Professor Medard from MIT)• Publications

– 1 edited book– 1 book chapter (several on the own book projects)– 2 journal (one submitted)– 12 conference