04/20/23 1

Title: An Adaptive Queue Management Title: An Adaptive Queue Management Method for Congestion Avoidance in Method for Congestion Avoidance in TCP/IP NetworksTCP/IP Networks

Presented By:Presented By:

Frank Posluszny

Vishal Phirke

Matt Hartling

04/20/23 2

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 3

Background (1)Background (1)

Goals:

Show drawbacks of RED with ECN

Propose new AQM: Random Early Adaptive Detection

04/20/23 4

Background (2)Background (2)

TCP congestion controlCongestion Control vs. AvoidanceREDECN

04/20/23 5

ECN:Binary feedback schemeRouter sets a bit in packet to “mark”

instead of dropACK mirrors the marking back to

receiver

Background (3)Background (3)

04/20/23 6

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 7

What’s Power??

Throughput optimized N/W

-Great throughput- Takes 15minutes to view a web page.

Delay optimized N/W

-Low Delays – But the web page is missing a lot of information.….

Throughput Delay

Power = Throughput

Response Time

04/20/23 8

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 9

Simulation Topology

Bottleneck

Queue Size = 60 pktsPkt Size = 512 bytes

MINth= 15

MAXth= 45

04/20/23 10

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 11

Weakness of RED - Motivation

04/20/23 12

Weakness of RED - Motivation

04/20/23 13

Weakness of RED - Motivation 10 flows

04/20/23 14

Weakness of RED - Motivation 60 flows

04/20/23 15

Weakness of RED - Motivation 20 flows

04/20/23 16

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 17

Random Early Adaptive Detection

Exponentially Weighted Moving Averages

Avgt+1 = (1-wq) avgt + wq qt

Slt+1 = (1-wsl) slt + wsl (avgt+1 – avgt)

Instantaneous queueOld weighted average

Instantaneous slopeOld weighted slope

04/20/23 18

Random Early Adaptive Detection

At each change of MIN

(MAX + MIN)

2 level =

if(level > buffer * 0.52)

p = p + INC

if(level < buffer * 0.48)

p = p - DEC

INC = 0.02

DEC = 0.002

04/20/23 19

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 20

Fig 5: Throughput Vs. DelayFig 5: Throughput Vs. Delay

04/20/23 21

READ Vs. RED (1)READ Vs. RED (1)

RED:Lower Drop probability = Higher

Throughput & Higher DelayHigher Drop probability = Lower

Delay & Lower ThroughputREAD:Always Lower Delay and Higher

Throughput

04/20/23 22

Fig 6: Power (alpha=1)Fig 6: Power (alpha=1)

04/20/23 23

Fig 7: Power (alpha = 2)Fig 7: Power (alpha = 2)

04/20/23 24

READ Vs. RED (2)READ Vs. RED (2)

RED:Performance varies with maxp and

number of flowsPerforms worse than Drop Tail under

certain conditionsREAD:Always performs better than RED

and Drop Tail

04/20/23 25

Table 1: Throughput For Mixed TrafficTable 1: Throughput For Mixed Traffic

04/20/23 26

Fig 8 & 9: Adaptation to Changes in Network Fig 8 & 9: Adaptation to Changes in Network ConditionsConditions

04/20/23 27

READ Vs. RED (3)READ Vs. RED (3)

RED: Large variation in instantaneous and

average queue size Large variation in marking probability Marking probability varies with queue sizeREAD: Less variation in marking probability and

queue size Large, periodic fluctuations

04/20/23 28

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 29

Fig 10: READ TuningFig 10: READ Tuning

04/20/23 30

Outline

Background

Network Power Simulation Topology

Weakness of RED - Motivation

Algorithm

Conclusions and Future Work

Simulations & Comparisons RED Vs READ

READ Tuning

04/20/23 31

Conclusions and Future WorkConclusions and Future Work

Conclusions: RED can fail & too aggressive READ – reliable CA; higher power levels

Current & Future Work: Examine different increase/decrease

algorithms READ with different Network Topologies