Weighted Fairness Performance Scenarios - IEEE 802 · Weighted Fairness Performance Scenarios Khaled Amer AmerNet IEEE 802.17 Plenary Meeting Vancouver, BC, Canada July 2002

Weighted FairnessPerformance Scenarios

Khaled AmerAmerNet

IEEE 802.17 Plenary MeetingVancouver, BC, Canada

July 2002

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

2

Background

• RAH/FAH adhoc decided that it would bebeneficial to have an annex with scenariosthat may have fairness/performance issues

• Only some of these scenarios are presentedhere, others are in annex J

• Can be used as implementation guidelines

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

3

Objectives

• Identify scenarios that may havefairness/performance implications

• Show the ideal target for a fairnessmechanism

• Show the effect of not having a fairnessmechanism as well as potential fairnessproblems

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

4

Convention Used

Ideal Target

Potential problem

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

5

Parking Lot Scenario #1

s1 s3 s4 s5 s6 s7 s8s2

flow[1,5]

flow[2,5]

flow[3,5]

flow[4,5]

25%

25%0%

0%

25%0%

25%100%

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

6

Parking Lot Scenario #1 ...

• Concern:

– Station #1 consumes total BW

• Solution:

– Throttle upstream node using fairnesscontrol messages

• Applicable:

– Addressed by single choke and multi-choke fairness algorithms

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

7

Parking Lot Scenarios #2

s1 s3 s4 s5 s6 s7 s8s2

flow[1,5]

flow[2,5]

flow[3,5]

flow[4,5]

25%

25%50%

25%

25%12.5%

25%12.5%

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

8

Parking Lot Scenario #2 ...

• Concern:

– Station #1 is overly restricted

• Solution:

– Have fairness algorithm divide BWevenly between upstream stations

• Applicable:

– Addressed by single choke and multi-choke algorithms

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

9

Parallel Parking Lot

s1 s3 s5 s6 s7 s8s2

flow[2,6]

flow[3,6]

flow[4,6]

flow[5,6]flow[2,3]

25%

25%

25%

s4

75%

25%

0%

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

10

Parallel Parking Lot ...• Concern:

– Flow [2,3] is restricted by congestion on link [5,6]

• Solution:

– Have fairness algorithm provide congestion info andhop-count to congestion point

– Provide support for VDQ for various hop-countdistances

• Applicable:

– When hop count is not reported

– Addressed by single choke and multi-chokealgorithms with VDQ

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

11

Multi-flow Parking Lot

• Illustrates support of weighted fairnesswith aggregate flows

s1 s3 s5 s6 s7 s8s2

flow[1,5]

flow[2,5]

flow[3,5]

flow[4,5]

s4

flow[4,6]

flow[4,7]

flow[4,8]

25%

25%

6.25%

14%25%14%

25%14%

14%6.25%

14%6.25%

14%6.25%

14%

56%

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

12

Multi-flow Parking Lot ...

• Concern:

– Flows from node 4 get more than 25% BW whichconflicts with source-based fairness

• Solution:

– Have fairness algorithm provide source-based fairness

• Applicable:

– Addressed by single choke and multi-choke algorithms

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

13

Dual Exit Parking Lot

• Illustrate effect of having multiple chokepoints on the ring

s1 s3 s5 s6 s7 s8s2

flow[2,5]

flow[3,5]

s4

flow[2,7]

flow[3,7]

flow[4,7]

flow[5,7]

flow[6,7]

20%

20%

20%

20%

40%

20%0%

20%

20%

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

14

Dual Exit Parking Lot ...

• Concern:

– Flows traversing through link [4,5] are overlythrottled because they only observe congestion onlink [6,7]

• Solution:

– All stations should be made aware of all chokepoints and not just the worst one

• Applicable:

– Problem occurs with single choke fairness algorithm

– Solved using multi-choke fairness algorithm

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

15

Choked high/low BW pair

• Illustrates potential for oscillations

BA

s1 s3 s5 s6 s7 s8s2 s4

flow[1,8]95% (steady state)

flow[5,8] 5%50% (average worst case)

choke

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

16

Choked high/low BW pair …(Ramp time dominates prop time)

A'sallowed rate

A is periodically shutoff by B and has toramp all the way up

Time

BW

use

d by

A

5%

95%

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

17

Choked high/low BW pair ...

• Concern:

– Flows from node 1 are unnecessarily throttled in acyclical fashion which reduces BW utilization

• Solution:

– A solution is to have fairness algorithm provideinformation about the whole ring, and react based onring conditions

• Applicable:

– Problem occurs with aggressive mode

– Worst case occurs with two stations far apart

IEEE 802.17 Plenary MeetingVancouver, BC / July 2002 ka_scen_02.pdf

Khaled Amer

18

Conclusions

• These scenarios (and others) arerecommended by the RAH/FAH to beadded as informative text in Annex J tohelp understand the implications ofvarious fairness algorithms

• Behavior of fairness algorithms need tobe simulated to demonstrate how theyoperate in these scenarios