Network Coding-Aware Queue Network Coding Aware Queue Management - - PowerPoint PPT Presentation

Network Coding-Aware Queue Network Coding Aware Queue Management for Unicast Flows over Coded Wireless Networks Coded Wireless Networks

Hülya Seferoğlu, Athina Markopoulou

University of California Irvine University of California, Irvine

Wireless Mesh Networks Wireless Mesh Networks

• Y. Wu, P. A. Chou, S. Y. Kung, “Information exchange in wireless network

coding and physical layer broadcast” CISS ’05 coding and physical layer broadcast , CISS 05.

• S. Katti, H. Rahul, W. Hu, D. Katabi, M. Medard, J. Crowcroft "XORs In

The Air: Practical Wireless Network Coding, (COPE)“, ToN ’08.

• Throughput increases by mixing packets
• Throughput increases by mixing packets

Gateway Intermediate Node

Internet

Intermediate Node

One-hop network coding One hop network coding

a3 a2 a1

Output queue Output queue Output queue

A B

a1 b1 b3 b2 b1 b1 I a1 b1 a1 + b1

D C

b1 a1

Virtual queue Virtual queue

Motivation Motivation

• Problem:
• TCP over COPE does not fully exploit the NC potential
• Intuition:

N t h di t iti d t TCP b ti

• Not enough coding opportunities due to TCP burstiness
• Coded flows do not compete for resources

TCP

x1 x2

TCP

A B

I

x1

a1

x2

a1 b1 b1 I

D C

max{x1, x2}

b1

C

Packets in the buffer☺

?

Motivation Motivation

• Problem:
• TCP over COPE does not fully exploit the NC potential
• Intuition:

N t h di t iti d t TCP b ti

• A Possible Solution:
• Not enough coding opportunities due to TCP burstiness
• Coded flows do not compete for resources
• A Possible Solution:
• Artificially delay packets at intermediate nodes
• Y. Huang, M. Ghaderi, D. Towsley, and W. Gong, “TCP performance in coded wireless

mesh networks,” SECON ‘08.

• Throughput increases with small delay, but decreases with large

delay

• Optimal delay depends on the network topology and the

b k d t ffi d h ti background traffic, and may change over time

• Not practical

Motivation Motivation

• Problem:
• TCP over COPE does not fully exploit the NC potential
• Intuition:

N t h di t iti d t TCP b ti

• Not enough coding opportunities due to TCP burstiness
• Coded flows do not compete for resources
• Proposed Solution:
• Network Coding-Aware Queue Management (NCAQM)
• No changes to TCP and MAC
• No changes to TCP and MAC
• Formulate network utility maximization (NUM) problem
• TCP+NCAQM doubles the network coding benefit of

TCP COPE TCP+COPE

Previous Work

I i N k C di Intra-session Network Coding

• Minimum cost multicast for wired and wireless:
• D. S. Lun, N. Ratnakar, M. Medard, R. Koetter, D. R. Karger, T. Ho,
• E. Ahmed, and F. Zhao, “Minimum-cost multicast over coded packet

networks,” ToIT'06.

• L. Chen, T. Ho, S. Low, M. Chiang, and J. C. Doyle, “Optimization

based rate control for multicast with network coding,” Infocom'07.

• Minimum cost unicast with for wireless:
• B. Radunovic, C. Gkantsidis, P. Key, P. Rodriguez, and W. Hu,

“Toward Practical Opportunistic Routing with Intra-session N k C di f M h N k ” T N’09 Network Coding for Mesh Networks,” ToN’09.

Previous Work

I i N k C di Inter-session Network Coding

• Optimal Scheduling and Routing:

p g g

• P. Chaporkar and A. Proutiere, “Adaptive network coding and

scheduling for maximizing througput in wireless networks,” Mobicom'07.

• S. Sengupta, S. Rayanchu, and S. Banarjee, “An Analysis of

Wireless Network Coding for Unicast Sessions: The Case for Coding-Aware Routing,” Infocom'07.

E ffi i t t k di

• Energy efficient network coding:
• T. Cui, L. Chen, and T. Ho, “Energy Efficient Opportunistic Network

Coding for Wireless Networks,” Infocom'08.

• End2end pairwise network coding:
• A. Khreishah, C. C. Wang, and N. B. Shroff, “Cross-layer
• ptimization for wireless multihop networks with pairwise

i i k di ” J C'09 intersession network coding,” JSAC'09.

Motivation Motivation

• Proposed Solution:
• Proposed Solution:
• Network Coding-Aware Queue Management (NCAQM)
• No changes to TCP and MAC

F l k ili i i i (NUM) bl

• Formulate network utility maximization (NUM) problem
• TCP+NCAQM doubles the network coding benefit of TCP+COPE
• Our work in perspective:
• Multiple unicast flows over wireless with given network coding

scheme and pre-determined routing paths scheme and pre determined routing paths

• Connection between optimization and protocol design
• Intuition for practical implementation

Outline Outline

• Introduction
• Network Utility Maximization (NUM)
• Network Coding-Aware Queue Management (NCAQM)
• Performance Evaluation
• Extensions & Summary

Network utility maximization y

Formulation

x1 x2

U(x1) U(x2) A h R H t x U

k s k s S s s s x

∀ ≤

∑ ∑

∈

} { ) ( max

, , τ α

Optimize total utility

A B

I

x1

a1

x2

a1 b1 b1 P i S s A h R x H t s

A h J h s S s K k k s h h h K k s k s h k s h S s

k h h k

∈ ∈ ∀ = ∈ ∀ ≤

∑ ∑ ∑

∈ ∈ ∈ ∈ ∈

, , 1 , } { max . .

)| ( | , , ,

α τ α I

D C

max{x1, x2}

b1 A Cq

C h h A h J h S s K k

q k h

⊆ ∀ ≤

∑

∈ ∈ ∈ ∈

,

)| ( |

γ τ

Capacity constraint Flow Conservation Interference Interference

Network utility maximization y

Solution I:

∑

k s k s k s

H

∑ ∑

∈

=

k

k s h S s k s h s k s h k s h m

m t s m x H 1 . . max

, , , , α

} { max

, , s k s h k s h S s

x H

k

α

∈

≡

∈ k S s

x U

∑

) ( max A h R m x H t s x U

h h K k S s k s h s k s h k s h S s s s x

h k

∈ ∀ ≤

∑∑ ∑

∈ ∈ ∈

, ) ( . . ) ( max

* , , , , ,

τ α

τ α

A C P i S s

h A h J h s S s K k k s h

k h

⊆ ∀ ≤ ∈ ∈ ∀ =

∑ ∑ ∑

∈ ∈ ∈

, , , 1

)| ( | ,

γ τ α A Cq

C h h

q

⊆ ∀ ≤

∑

∈

, γ τ

Network utility maximization y

Solution II:

⎟ ⎞ ⎜ ⎛

Rate

⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ =

∑ ∑

∈ ∈ ∈ − A h S s K k k s h k s h k s h h s s

k h

m H q U x

| * , , , 1

) ( ) ' ( α

+

⎪ ⎫ ⎪ ⎧ ⎤ ⎡

Rate Control Qu u

∑ ∑

k k k * ∈ ∈

⎪ ⎭ ⎪ ⎬ ⎫ ⎪ ⎩ ⎪ ⎨ ⎧ ⎥ ⎦ ⎤ ⎢ ⎣ ⎡ − + = +

∑∑

h h K k S s s k s h k s h k s h t h h

R x m H c t q t q

k k

τ α

* , , ,

) ( ) ( ) 1 (

Queue Size

∑ ∑ ∑ ∑

∈ ∈ ∈ ∈ ∈ ∈

∈ ∀ =

A h J h S s K k s k s h A h J h S s K k k s h k s h k s h h

k h k h

P i t s m H q

)| ( | , )| ( | , * , ,

, 1 . . ) ( min α α

α

Traffic Splitting

∈ ∈ ∈ A h J h S s K k

k h

)| ( |

R q

A h h h h

∑

∈

max τ

τ

Scheduling

A Cq

C h h

q

⊆ ∀ ≤

∑

∈

, γ τ

Scheduling

Network Coding-Aware Queue Management

(NCAQM)

P t l difi ti i i ki th ti l l ti

Implementation Summary

Protocol modifications, mimicking the optimal solution

Implementation Summary

Queue management (NCAQM)

• Network coding
• Packet dropping

( Q ) pp g TCP No change (TCP-SACK) MAC No change (802 11) MAC No change (802.11)

Minimal and intuitive

NCAQM Q

Maintaining queues and packet transmission

+

⎫ ⎧

{ }

+ ∈ ∈

⎪ ⎭ ⎪ ⎬ ⎫ ⎪ ⎩ ⎪ ⎨ ⎧ ⎥ ⎦ ⎤ ⎢ ⎣ ⎡ − + = +

∑

h h K k s k s h k s h S s t h h

R x H c t q t q

k k

τ α ,

,

max ) ( ) 1 (

Queue Size

A B

i

Q

Modification I

• Qi is the output queue at node i
• Store network coded packets (when an

I

i

Q

• Store network coded packets (when an
• pportunity arises) instead of uncoded

packets

• Keep track of hypearc queues

D C

• Keep track of hypearc queues
• Estimate traffic splitting parameters
• Packet scheduling is according to

FIFO FIFO queue

NCAQM Q

Rate control and packet dropping

l

1 −

⎞ ⎛ S f k d d i

Modification II

Optimal Rate Control

) ( ) ( ∈

⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ = ∑

s

P i h s i h s

q x Sum of network coded queue sizes across all nodes on the path

Modification II

• Upon congestion, compare Qi

s for all

flows s. Drop an uncoded packet from the largest flow

A B

i

Q

x

from the largest flow

• How to calculate Qi

s ?

• Determine hyperarc queues that

fl s is d min tin (h s th I flow s is dominating (has the largest number of packets)

• Sum the number of packets of

flow s over these hyperarc

D C

flow s over these hyperarc queues

NCAQM

P bl m:

Q

Implementation Summary

• Problem:
• TCP over COPE does not fully exploit the NC potential
• Intuition:
• Flows coded together do not compete for resources
• Not enough coding opportunities due to TCP burstiness
• Modifications
• Store network coded packets (when an opportunity arises)

instead of uncoded packets instead of uncoded packets.

• Compare Qi

s for all s. Drop an uncoded packet from the

“largest” flow.

Outline Outline

• Introduction
• Network Utility Maximization (NUM)
• Network Coding-Aware Queue Management (NCAQM)
• Performance Evaluation
• Extensions & Summary

Performance evaluation

Scenarios

[Glomosim + NC]

A & B Topology X Topology p gy l Cross Topology Grid Topology

Performance evaluation

Throughput improvement compared to noNC

TCP+COPE (%) TCP+NCAQM (%) Optimal (%) (%) (%) (%) A & B 12 27 33 Cross 16 31 60 X 10 22 33 X 10 22 33 Grid 8 19

• TCP+NCAQM doubles the improvement of TCP+COPE

Performance evaluation

Throughput improvement vs buffer size

Extensions

Multi-hop network coding

• Network utility maximization problem is extended for

multi-hop network coding Di t ib t d l ti d i d

• Distributed solutions are derived
• Only traffic splitting part changes
• In practice, traffic splitting parameter is estimated
• NCAQM is directly applied to multi-hop network coding

Summary Summary

• Proposed queue management schemes to improve the

p q g p performance of TCP over coded wireless networks

• Formulated network utility maximization problem and
• Formulated network utility maximization problem and

proposed a distributed solution D si d NCAQM s h m mimi ki th st t f th

• Designed NCAQM scheme, mimicking the structure of the
• ptimal solution. No changes TCP and MAC.

l h h P d bl h

• Simulations show that TCP+NCAQM doubles the

improvement of TCP+COPE as compared to noNC.

Thank you! y

hseferog @uci.edu http://newport.eecs.uci.edu/~hseferog/