Achieving Throughput Fairness in Achieving Throughput Fairness in Wireless Mesh Network Based on Wireless Mesh Network Based on IEEE 802.11IEEE 802.11Janghwan Lee and Ikjun YeomJanghwan Lee and Ikjun Yeom

Division of Computer ScienceDivision of Computer Science

KAISTKAISTjlee@cnlab.kaist.ac.krjlee@cnlab.kaist.ac.kr yeom@cs.kaist.ac.kryeom@cs.kaist.ac.kr

Wireless Mesh NetworkWireless Mesh Network

‣Multi-hop wireless infrastructureMulti-hop wireless infrastructure‣Uses IEEE 802.11Uses IEEE 802.11

Residential user

or small business

Internet

Gateway node

Router node

Client node

Fairness ProblemFairness Problem

Unfair bandwidth sharing among flows with Unfair bandwidth sharing among flows with different hop distancedifferent hop distance

Related WorkRelated Work

Centralized scheme [Gamboriza]Centralized scheme [Gamboriza]

Every node in the network knows about Every node in the network knows about global topology and offered load.global topology and offered load.

Calculate proper ingress rate at each node.Calculate proper ingress rate at each node.

Hard to know offered load.Hard to know offered load.

Not scalable.Not scalable.

Related WorkRelated Work

Queueing SchemeQueueing Scheme

Per flow queueing [Jun]Per flow queueing [Jun]

Limit queue share [Nandiraju]Limit queue share [Nandiraju]

Cannot resolve MAC layer contentionCannot resolve MAC layer contention

Wasting bandwidthWasting bandwidth

Cannot apply to the case when queue is not Cannot apply to the case when queue is not occupied enoughoccupied enough

Problem StatementProblem Statement

Support the same bandwidth to every leaf node with distributed Support the same bandwidth to every leaf node with distributed algorithm.algorithm.

uplinkuplink

Assume multi-radio, multi-channelAssume multi-radio, multi-channel

To make a model simpleTo make a model simple

Separate uplink and downlink channelSeparate uplink and downlink channel

Eliminate propagation of interferenceEliminate propagation of interference

Independent collision domainIndependent collision domain

Common in [Brezezinski], [Raniwala]Common in [Brezezinski], [Raniwala]

Find proper weight Find proper weight ff(p,i)(p,i) that satisfies that satisfies ccii==ccjj for all for all i,ji,j

Proposed SchemeProposed Scheme

Node Weight EstimationNode Weight EstimationNeed to know the number of active nodesNeed to know the number of active nodes

Just counting the number of nodes with timeout is too naiveJust counting the number of nodes with timeout is too naive

Hard to adjust timeout value because of highly dynamic networkHard to adjust timeout value because of highly dynamic network

We can estimate the weight by dividing the aggregate throughput by We can estimate the weight by dividing the aggregate throughput by the average throughput of leaf nodesthe average throughput of leaf nodes

How to know How to know LLii??

Node Number Node Number EstimationEstimation

Leaf nodes piggyback their sending rate on Leaf nodes piggyback their sending rate on packetspackets

Intermediate nodes calculate average value Intermediate nodes calculate average value from leaf ratefrom leaf rate

Need to compensate biasNeed to compensate bias

Weighted SchedulingWeighted Scheduling““Achieving efficient channel utilization and weighted fairness for data Achieving efficient channel utilization and weighted fairness for data communications in IEEE 802.11 WLAN under DCF”, In Proc. of IWQOS communications in IEEE 802.11 WLAN under DCF”, In Proc. of IWQOS 20022002

Differentiate nodes using collision modelDifferentiate nodes using collision model

Has a problem in multi-hop networkHas a problem in multi-hop network

Weighted SchedulingWeighted SchedulingFrom the model in [Qiao] and [Bianchi], the probability that node i From the model in [Qiao] and [Bianchi], the probability that node i attempts to transmit on given slot as,attempts to transmit on given slot as,

The probability that node i successfully transmits on a given slotThe probability that node i successfully transmits on a given slot,,

To satisfy the weight of node i and j, fTo satisfy the weight of node i and j, fii and f and fjj,,

Weighted SchedulingWeighted Scheduling

With exponential backoff,With exponential backoff,

Unfortunately, we cannot get the solution of Unfortunately, we cannot get the solution of equations with closed form.equations with closed form.

Numerical solutionNumerical solution

Weighted SchedulingWeighted SchedulingContention window size according to weighted Contention window size according to weighted scheduling model when the base contention scheduling model when the base contention

window is 30window is 30

Weighted SchedulingWeighted Scheduling

Normalized throughput with changing the Normalized throughput with changing the weight from 2 to 5 using different modelsweight from 2 to 5 using different models

Simulation ResultSimulation Result

Simulation TopologySimulation Topology

Simulation ResultSimulation ResultAt 20 sec

stops at 60

At 40 sec

At 60 sec

Initial

Weight estimation and weighted schdulingWeight estimation and weighted schduling

Simulation ResultSimulation ResultTCP Throughput of each node (sorted in ascending TCP Throughput of each node (sorted in ascending

order)order)

Aggregate throughput of proposed Aggregate throughput of proposed scheme achieves 97% of the IEEE scheme achieves 97% of the IEEE

802.11 throughout simulations802.11 throughout simulations

Simulation ResultSimulation Result

UDP throughput of each UDP throughput of each node at different load node at different load (sorted in ascending (sorted in ascending

order)order)

ConclusionConclusion

We proposed a scheme to realize throughput We proposed a scheme to realize throughput fairness in wireless mesh network.fairness in wireless mesh network.

Our scheme performed well without significant Our scheme performed well without significant loss of aggregate throughputloss of aggregate throughput

IssuesIssues

Are the assumptions about topology and Are the assumptions about topology and channel environment reasonable?channel environment reasonable?

Cannot apply to the single channel or Cannot apply to the single channel or unplanned networkunplanned network

ReferencesReferences

[Gambiroza] V. Gambiroza et al. “End-to-end performance and fairness in multihop wireless [Gambiroza] V. Gambiroza et al. “End-to-end performance and fairness in multihop wireless backhaul networks.”, in proc. of Mobicom ’04backhaul networks.”, in proc. of Mobicom ’04

[Nandiraju] N. S. Nandiraju et al., “A novel queue management mechanism for improving [Nandiraju] N. S. Nandiraju et al., “A novel queue management mechanism for improving performance of multihop flows in IEEE 802.11s based mesh networks” In Proc. of IPCCC 2006, performance of multihop flows in IEEE 802.11s based mesh networks” In Proc. of IPCCC 2006, April 2006.April 2006.

[Jun] J. Jun and M. L. Sichitiu, “Fairness and QoS in multihop wireless network.” In Proc. of VTC [Jun] J. Jun and M. L. Sichitiu, “Fairness and QoS in multihop wireless network.” In Proc. of VTC 2003-Fall2003-Fall

[Qiao] D. Qiao et al., “Achieving efficient channel utilization and weighted fairness for data [Qiao] D. Qiao et al., “Achieving efficient channel utilization and weighted fairness for data communications in IEEE 802.11 WLAN under DCF”, In Proc. of IWQOS 2002communications in IEEE 802.11 WLAN under DCF”, In Proc. of IWQOS 2002

[Bianchi] G. Bianchi, “Performance anlysis of the IEEE 802.11 distributed coordination [Bianchi] G. Bianchi, “Performance anlysis of the IEEE 802.11 distributed coordination function.”, JSAC March, 2000.function.”, JSAC March, 2000.