Presented by: Frank Posluszny Vishal Phirke

04/21/23 1

Adaptive Adaptive REDRED::An Algorithm for Increasing the An Algorithm for Increasing the Robustness of Robustness of REDRED’s Active Queue ’s Active Queue ManagementManagementororHow I learned to stop worrying and love How I learned to stop worrying and love REDRED

Presented by:Frank Posluszny

Vishal Phirke

04/21/23 2

Introduction

Adaptive RED Algorithm.

Parameters and their values

Simulations.

Delay-Throughput Tradeoff.

Conclusions.

Background and Related Work

Pre-resultsPre-results

Metrics and Scenarios

04/21/23 3

Introduction - 1Introduction - 1

Who are they authors?– Sally Floyd (original RED author)– Ramakrishna Gummadi (CS grad - intern)– Scott Shenker (works with Sally Floyd)

04/21/23 4

Introduction - 2Introduction - 2

Goals:– People want a guaranteed delay, which

RED can’t do without constantly adjusting the parameters

– “Our goal… is to solve this problem with minimal changes to the overall RED algorithm.”

04/21/23 5

Introduction



Simulations.


Conclusions.




04/21/23 6

Background & Related Work - 1Background & Related Work - 1

Quick review of RED– Try to maintain queue size under a

threshold, assuming that as we get closer to that threshold then congestion will start to occur

– Once we “foresee” congestion, drop with an increasing probability

04/21/23 7


Problems, problems everywhere…– Tuning RED for Web Traffic– A number of papers point out problems

with oscillations in the instantaneous queue size (Misra et at., Hollot et al., Firoiu et al.)

– Average queuing delay– Throughput

04/21/23 8


Suggested fixes…– Jacobson (how to set wq)

– Ziegler (tighter bound for aveq)

– Feng (adapt maxp)

– AVG (keep queue size small, token bucket)

– SRED (estimate #of active flows)– DRED (keep queue size near a threshold)

04/21/23 9

Introduction



Simulations.


Conclusions.




04/21/23 10

Metrics & Scenarios - 1Metrics & Scenarios - 1

the NS network simulator is used for all tests/scenarios

router-based metrics vs. user-based worst-case is not their concern not looking at queue length

oscillations directly

04/21/23 11

Metrics & Scenarios - 2Metrics & Scenarios - 2

Wide range or traffic scenarios– range of workloads (long vs. short lived)– levels of statistical multiplexing– levels of congestion– reverse traffic– with & without ECN– large window advertisements– different packet sizes

04/21/23 12

Introduction



Simulations.


Conclusions.




04/21/23 13

Delay-Utilization tradeoff with RED. wq = 0.002

Delay-Utilization tradeoff with RED. wq = 0.00026

04/21/23 17

Introduction



Simulations.


Conclusions.




04/21/23 18

Adaptive RED Algorithm

• MAXp is adapted to keep AVGq around (MINth +MAXth)/2

•AIMD is used for Adapting MAXp

•Adaptation is slow – 0.5sec interval

•MAXp bounded between [0.01,0.5]

MINth

MAXth

AVGq

Packet Queue

04/21/23 19

Adaptive RED Algorithm

Every interval seconds(0.5sec):

If(AVG > Target and MAXp < 0.5)

MAXp = MAXp +

Else if(AVG < Target and MAXp > 0.01)

MAXp = MAXp *

= Min(0.01, MAXp/4) = 0.9

04/21/23 20

Introduction



Simulations.


Conclusions.




04/21/23 21

MAXp Range [0.01, 0.5]

Upper bound 0.5

Not trying to optimize for packet drop rates more than 50%.

Gentle RED

MINth MAXth 2MAXth

MAXp

1

Lower bound 0.01

No one will object lower delays

Limits the MAXp Range – Important as Adaptation is slow (0.5 sec interval)

04/21/23 22

Values of Increment And decrement

Target Range

(AVG – MINth)

(MAXth – MINth) p = MAXp

AVG1 = MINth + (MAXth – MINth) p

MAXp

AVG2 = MINth + (MAXth – MINth) p

MAXp +

= min(0.01, MAXp/4) = 0.9

Target Range > AVG2 – AVG1

04/21/23 23

MAXth & Wq

MAXth = 3 * MINth - As per latest recommendation of Sally Floyd. (They don’t follow it in their simulations.)

Wq gives a Time Constant in terms of packet arrival rate for AVG queue to adapt. ( -1/ln(1-Wq) ) - Original RED

Since it is in terms of packet arrival rate, should be dependent on link capacity C.

C = -1/ln(1-Wq) Wq = 1 – exp(-1/C)

04/21/23 24

Introduction



Simulations.


Conclusions.




04/21/23 25

RED, one-way long-lived traffic, Wq= 0.002

100 long-lived flows, 250ms RTT, MINth=20, MAXth=80

04/21/23 26

Adaptive-RED, one way long-lived traffic Wq = 0.00027

100 long-lived flows, 250ms RTT, MINth=20, MAXth=80

04/21/23 27

RED, two flows, Wq = 0.002 (Large Wq)

2 TCP flows, 1st start at time 0, 2nd at 2.5sec

04/21/23 28

RED, automatic setting for Wq, 0.00027


04/21/23 29

RED, Wq too small, 0.0001


04/21/23 30

Introduction



Simulations.


Conclusions.




04/21/23 31

AVGq

Delay-Throughput Tradeoff

AVGq = MINth + MAXth

2

MAXth = 3 * MINth

AVGq = 2* MINth

Delaytarget * C = 2 * MINth

MINth = Delaytarget * C

2

04/21/23 32

Introduction



Simulations.


Conclusions.




04/21/23 33

Conclusions

Reduces RED’s parameter sensitivity.

Network operators can configure delay by

using proper value for MINth.

04/21/23 34

Documents

Presented by: Frank Posluszny Vishal Phirke