A Hybrid Approach to Effort- Limited Fair (ELF) Scheduling for 802.11 By David Matsumoto June 20 th, 2003.

A Hybrid Approach to Effort-A Hybrid Approach to Effort-Limited Fair (ELF) Scheduling Limited Fair (ELF) Scheduling

for 802.11for 802.11

By David MatsumotoBy David Matsumoto

June 20June 20thth, 2003, 2003

AgendaAgenda

• IntroductionIntroduction

• Effort-Limited Fair (ELF) SchedulingEffort-Limited Fair (ELF) Scheduling

• The ThesisThe Thesis

• Integrate ELF into 802.11Integrate ELF into 802.11

• Hybridize ELFHybridize ELF

• Simulation setup & resultsSimulation setup & results

• Possible future research & recommendationsPossible future research & recommendations

• ConclusionConclusion

IntroductionIntroduction

applicationtransportnetworkdata linkphysical

applicationtransportnetworkdata linkphysical

applicationtransportnetworkdata linkphysical

• Applications on a Applications on a networknetwork– Different requirements Different requirements

(bandwidth, timing, (bandwidth, timing, reliability)reliability)

• Network stackNetwork stack– Underlying protocols Underlying protocols

(TCP, UDP, IP, MAC, etc…)(TCP, UDP, IP, MAC, etc…)• Providing quality of Providing quality of

service (QoS)service (QoS)– Wired versus wirelessWired versus wireless

• Wireless: subject to Wireless: subject to capacity losscapacity loss

• ““Effort” spent does not Effort” spent does not always equal “outcome”always equal “outcome”

Earlier Work: Effort-Limited Fair Earlier Work: Effort-Limited Fair (ELF) Scheduling(ELF) Scheduling

• MotivationMotivation– Applications have poor performance due to link Applications have poor performance due to link

errorserrors– Link errors reduce useful link throughputLink errors reduce useful link throughput

• Solutions introduced: Swapping time slots, Adaptive Solutions introduced: Swapping time slots, Adaptive forward error controlforward error control

• Capacity loss is a fundamental reality for wireless Capacity loss is a fundamental reality for wireless schedulersschedulers..

• QuestionsQuestions– Can we support reservations amidst capacity loss? Can we support reservations amidst capacity loss? – How should we redistribute the remaining capacity How should we redistribute the remaining capacity

among flows? among flows?

Example: 50% Packet LossExample: 50% Packet Loss• Wireless cell capacity: 800 kilobits/secondWireless cell capacity: 800 kilobits/second

• WFQ scheduler serves two guaranteed WFQ scheduler serves two guaranteed flows & two best-effort flowsflows & two best-effort flows

ClientClient Expected Expected Rate Rate (kbit/s)(kbit/s)

Effort-fair Effort-fair rate rate (kbit/s)(kbit/s)

Preferable Preferable rate rate (kbit/s)(kbit/s)

AudioAudio 88 4 ×4 × 8 8 √√

VideoVideo 350350 175 175 ××

350 350 √√

FTP 1FTP 1 221221 110 110 2121

FTP 2FTP 2 221221 110 110 2121

Challenges in QoSChallenges in QoS

• Capacity loss should result in Capacity loss should result in throughput loss according to throughput loss according to administrative controlsadministrative controls

• What about location dependent errors?What about location dependent errors?– How should we regulate air-time? How should we regulate air-time?

• Conclusion: Conclusion: – We must have a way to balance We must have a way to balance fidelityfidelity

with with efficiencyefficiency. .

ELF Scheduling – Design ELF Scheduling – Design PrinciplesPrinciples• Wireless scheduler should be equivalent to Wireless scheduler should be equivalent to

wireline scheduler in error-free environmentwireline scheduler in error-free environment• Capacity loss suffered per flow should be Capacity loss suffered per flow should be

administratively configurableadministratively configurable• Administratively bound capacity lost due to Administratively bound capacity lost due to

location-dependent errorslocation-dependent errors• Unless configured otherwise, flows with the same Unless configured otherwise, flows with the same

error rate should experience the same capacity error rate should experience the same capacity lossloss

• Capacity unused by one flow should be Capacity unused by one flow should be distributed “fairly” among other flowsdistributed “fairly” among other flows

The ELF Scheduler The ELF Scheduler

Admission Control

Application Protocols

Packet Schedulerweights

UserRequests

AdministrativeControls

“next packet”

Link Layer

Power factors

Effort/outcome

ELF Power factorELF Power factor

• Main Idea:Main Idea:– Extend transmission time in a controlled fashionExtend transmission time in a controlled fashion– Allot a flow extra effort to meet its reservations, Allot a flow extra effort to meet its reservations,

up to some administrative bound up to some administrative bound Power factor Power factor

• Adjusted Weight (Adjusted Weight (AAii))– AAii = = min min ((WWii/1 – /1 – EEi i , , PPii X X WWii))

• Effective throughput (Effective throughput (TTii))– TTii = (( = ((AAii/∑/∑jj AAjj) X ) X BB) x (1 – ) x (1 – EEii))

• This effectively balances This effectively balances fidelityfidelity with with efficiencyefficiency. .

The ThesisThe Thesis

It is possible to integrate an Effort-It is possible to integrate an Effort-Limited Fair scheduler (ELF) into a Limited Fair scheduler (ELF) into a standard 802.11 implementation, and standard 802.11 implementation, and thus ensure sensible outcomes for flows thus ensure sensible outcomes for flows in response to unrecoverable capacity in response to unrecoverable capacity loss. loss.

Moreover, it is possible to improve on the Moreover, it is possible to improve on the efficiency of ELF within an 802.11 MAC by efficiency of ELF within an 802.11 MAC by using a hybrid approach to regulate using a hybrid approach to regulate stations within both a DCF & a PCF.stations within both a DCF & a PCF.

IEEE 802.11IEEE 802.11

• Standard for wireless Local Area Networks (LAN)Standard for wireless Local Area Networks (LAN)• Distributed Coordinator Function (DCF)Distributed Coordinator Function (DCF)

– Uses Carrier Sense Multiple Access with Collision Uses Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA)Avoidance (CSMA/CA)

• Point Coordinator Function (PCF)Point Coordinator Function (PCF)– Centralized, polling-based mechanism requiring a base Centralized, polling-based mechanism requiring a base

station (Point Coordinator)station (Point Coordinator)• DCF & PCF can co-existDCF & PCF can co-exist

• Contention period (CP) under control of DCFContention period (CP) under control of DCF• Contention-free period (CFP) under control of PCFContention-free period (CFP) under control of PCF

• Both methods use explicit AcksBoth methods use explicit Acks– Useful for PC tracking outcomeUseful for PC tracking outcome

802.11 PCF/DCF Alternation802.11 PCF/DCF Alternation

Integration of ELF scheduling Integration of ELF scheduling into 802.11into 802.11• Core ELF algorithm remains the sameCore ELF algorithm remains the same

– Choose “most deserving” flowChoose “most deserving” flow– Flow is allocated effort as a function of its power factorFlow is allocated effort as a function of its power factor

• Must have “effort” allocated to be chosenMust have “effort” allocated to be chosen– Separate best-effort flows from guaranteed flowsSeparate best-effort flows from guaranteed flows

• Update each flow per clock “tick” Update each flow per clock “tick”

• ELF scheduling becomes the policy for the PCFELF scheduling becomes the policy for the PCF– Departs from existing protocol specificationsDeparts from existing protocol specifications

• Records “outcome” via returned Acks (or NULL Records “outcome” via returned Acks (or NULL frames)frames)

• Charge a station for “effort” each time it is polled. Charge a station for “effort” each time it is polled. – Station charged even if packets are corrupted. Station charged even if packets are corrupted.

Extending ELF for DCFExtending ELF for DCF

• MotivationMotivation– DCF is more efficient in general, when traffic DCF is more efficient in general, when traffic

load isn’t high. load isn’t high. – Ideal: PCF could interject only when necessary Ideal: PCF could interject only when necessary

to bring flows into conformance with to bring flows into conformance with reservations.reservations.

• Design PrinciplesDesign Principles– DCF must remain completely distributedDCF must remain completely distributed– ELF-DCF should adhere to original ELF design ELF-DCF should adhere to original ELF design

principles. principles. – Integration of ELF-DCF/PCF should be seamlessIntegration of ELF-DCF/PCF should be seamless

ELF-DCF: Implementation ELF-DCF: Implementation DesignDesign• Continue using both Continue using both

“deserve” & “effort” to (self) “deserve” & “effort” to (self) regulate flowsregulate flows

• ““Deserve” carries over from Deserve” carries over from CFPCFP

• ““Effort” allocated based on Effort” allocated based on past CP historypast CP history– Cannot force station to Cannot force station to

expend effortexpend effort– Make use of power factor Make use of power factor – ““Effort” is conserved Effort” is conserved

(unused effort plugged into (unused effort plugged into CFP)CFP)

• No distinction between best-No distinction between best-effort versus guaranteed effort versus guaranteed flowsflows– Reduce overhead in beaconReduce overhead in beacon

ELF Scheduler

(PC)

Mobile Nodes

Send Beacon with ELFdata

Collectstatistics/Start CFP

Implementation for Implementation for ns-2ns-2

• Build on top of CMU Monarch group’s workBuild on top of CMU Monarch group’s work– Basic mobile node functionality with 802.11 DCF Basic mobile node functionality with 802.11 DCF

• Add semi-complete implementation of PCF. Add semi-complete implementation of PCF. – Beacon timerBeacon timer– Bi-directional pollingBi-directional polling– Poll list – integrated with ELFPoll list – integrated with ELF

• Implement Weighted Round Robin (WRR) instead of WFQImplement Weighted Round Robin (WRR) instead of WFQ– Measure throughput in frame slots (done for simplicity)Measure throughput in frame slots (done for simplicity)– Not concerned with complex based service characterizations Not concerned with complex based service characterizations

(e.g. delay, jitter guarantees)(e.g. delay, jitter guarantees)• Administratively configure breakdown of superframeAdministratively configure breakdown of superframe

– I.E. Percentage CFP/CPI.E. Percentage CFP/CP• Assume admissions control module exists that can set Assume admissions control module exists that can set

appropriate per-flow power factors. appropriate per-flow power factors.

Experimental setupExperimental setup

• One flow per stationOne flow per station– Two CBR flows & two FTP Two CBR flows & two FTP

flows (each with 25% weight)flows (each with 25% weight)• Flows chosen so aggregate Flows chosen so aggregate

throughput consumes the throughput consumes the entire link ~ 1.4 Mbit/sentire link ~ 1.4 Mbit/s– CBR flow ~ 450 Kbit/sCBR flow ~ 450 Kbit/s– FTP flow ~ 250 Kbits/s (no FTP flow ~ 250 Kbits/s (no

errors)errors)• Relevant variables:Relevant variables:

– Percentage CFP/CPPercentage CFP/CP– ELF-DCF BooleanELF-DCF Boolean– Power factorPower factor– Error rateError rate

• Error modelsError models– Uniform, Markov, real-world Uniform, Markov, real-world

TracesTraces

WLAN

Uniform – 50% (CBR-1 Error prone)Uniform – 50% (CBR-1 Error prone)

50% Error rate, 0% CFP, ELF-DCF disabled

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Time (s)

Th

rou

gh

pu

t (M

bit

/s)

CBR(1)

CBR(2)

TCP(3)

TCP(4)

Total

• CBR-1 performs poorly while other stations operate normally• No ELF intervention here.

Uniform – 50% (CBR-1 Error prone)Uniform – 50% (CBR-1 Error prone)

50% Error Rate, 100% CFP, ELF-DCF Disabled

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Time (s)

Th

rou

gh

pu

t (M

bit

/s)

CBR(1)

CBR(2)

TCP(3)

TCP(4)

Total

• ELF-PCF effectively restores throughput to CBR-1 at the expense of best-effort flows

Uniform – 50% (CBR-1 Error prone)Uniform – 50% (CBR-1 Error prone)

50% Error rate, 50% CFP, ELF-DCF Disabled

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Time (s)

Th

rou

gh

pu

t (M

bit

/s)

CBR(1)

CBR(2)

TCP(3)

TCP(4)

Total

• ELF-PCF is only partially effective at restoring throughput

Uniform – 50% (CBR-1 Error prone)Uniform – 50% (CBR-1 Error prone)

50% Error rate, 50% CFP, ELF-DCF Enabled

0

0.2

0.4

0.6

0.8

1

1.2

Time (s)

Th

rou

gh

pu

t (M

bit

/s)

CBR(1)

CBR(2)

TCP(3)

TCP(4)

Total

• ELF-DCF effectively works with ELF-PCF to meet ELF design principles

Walls trace – (CBR-1 Error Walls trace – (CBR-1 Error prone)prone)

Wall - 0% PCF, ELF-DCF Disabled

00.2

0.40.60.8

11.2

1.41.6

Time (s)

Th

rou

gh

pu

t (M

bit

/s)

CBR(1)

CBR(2)

TCP(3)

TCP(4)

Total

• No ELF intervention – other flows gain throughput as CBR-1 decreases

Walls trace – (CBR-1 Error Walls trace – (CBR-1 Error prone)prone)

Walls - 50% PCF, ELF-DCF Enabled

0

0.2

0.4

0.6

0.8

1

1.2

1.4

Time (s)

Th

rou

gh

pu

t (M

bit

/s)

CBR(1)

CBR(2)

TCP(3)

TCP(4)

Total

• ELF-PCF/DCF work together• Other flows share throughput loss, up to administrative bound• Capping (“deserve”) mechanism needed

Future Work & Future Work & RecommendationsRecommendations

• Future workFuture work– Provide a capping mechanism for flows that Provide a capping mechanism for flows that

accumulate large “deserve” valuesaccumulate large “deserve” values– Can we “encourage” stations to spend effort Can we “encourage” stations to spend effort

during a DCF (rather than just during a DCF (rather than just limitinglimiting the effort the effort allotted)?allotted)?

• RecommendationsRecommendations– Incorporate a “policing” mechanism into beaconsIncorporate a “policing” mechanism into beacons– Change 802.11 specifications on how stations are Change 802.11 specifications on how stations are

polledpolled– Allow polling of individual flowsAllow polling of individual flows

ConclusionsConclusions

• ELF scheduling can be implemented ELF scheduling can be implemented as the policy for a PCF in 802.11as the policy for a PCF in 802.11

• ELF-DCF/PCF provides a hybrid ELF-DCF/PCF provides a hybrid mechanism through we can achieve mechanism through we can achieve ELF’s design principlesELF’s design principles

• Peter Steenkiste – AdvisorPeter Steenkiste – Advisor

• David Eckhardt – ReaderDavid Eckhardt – Reader

• INI Staff & ProfessorsINI Staff & Professors

AcknowledgementsAcknowledgements

QuestionsQuestions