a comparison of load-based and queue-based active queue management algorithms

19
1 Minseok Kwon and Sonia Fahmy Department of Computer Sciences Purdue University {kwonm, fahmy}@cs.purdue.edu All our slides and papers are available at: http://www.cs.purdue.edu/~fahmy/ A Comparison of A Comparison of Load-based and Queue-based Load-based and Queue-based Active Queue Management Active Queue Management Algorithms Algorithms

Upload: tolla

Post on 30-Jan-2016

66 views

Category:

Documents


0 download

DESCRIPTION

A Comparison of Load-based and Queue-based Active Queue Management Algorithms. Minseok Kwon and Sonia Fahmy Department of Computer Sciences Purdue University {kwonm, fahmy}@cs.purdue.edu All our slides and papers are available at: http://www.cs.purdue.edu/~fahmy/. Congestion Control. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

1

Minseok Kwon and Sonia Fahmy

Department of Computer SciencesPurdue University

{kwonm, fahmy}@cs.purdue.eduAll our slides and papers are available at:

http://www.cs.purdue.edu/~fahmy/

A Comparison of A Comparison of Load-based and Queue-based Load-based and Queue-based

Active Queue Management Active Queue Management AlgorithmsAlgorithms

Page 2: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

2

Congestion Control

Active Queue Management (AQM)

TCP Congestion Control

Page 3: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

3

TCP Congestion Control

Slow-Start

Congestion Avoidance

Additive IncreaseMultiplicative Decrease (AIMD)

Timeout

new ssthresh = cwnd / 2

cwnd

time

ssthresh

1 TCP-Reno3 DupAck

Page 4: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

4

Why Active Queue Management?

• Controls average queue size.• Absorbs bursts without dropping

packets.• Prevents bias against bursty

connections.• Avoids global synchronization of TCP.• Reduces the number of timeouts in TCP.• Punishes misbehaving flows.

Page 5: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

5

Queue-based Active Queue Management• RED [Floyd and Jacobson, 1993]

• Drops packets probabilistically in proportion to a long term average queue length (buffer occupancy).

• SRED [Ott et al., 1999]• The packet loss probability is proportional to the

instantaneous buffer occupancy and the estimated number of active flows.

• FRED [Lin and Morris, 1997]• Imposes on each flow a loss probability proportional to the

flow average and instantaneous buffer occupancy.

• BLUE [Feng et al., 2001]• Increments the packet drop probability when packet loss

occurs and decrements the packet drop probability if the link is idle.

Page 6: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

6

Load-based Active Queue Management

• REM [Athuraliya et al., 1999]• Congestion measure (price) is computed proportionally to the

difference between input rate and output rate and current buffer occupancy at router. Source rate is computed inversely proportional to the congestion measure. Thus, the source reaches a globally optimal equilibrium.

• AVQ [Kunniyur and Srikant, 2001]• Maintains a virtual queue. When the virtual queue overflows,

packets in a real queue are marked/dropped. The virtual capacity is modified such that total flows achieve a desired utilization of the link.

• PI controller [Hollot et al., 2001]• Queue length slope determines packet drop probability and the

queue is regulated to the desired queue length.

Page 7: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

7

Queue-based AQM: RED

Mark with PLinearly increasingFrom 0 to Pmax

No droppingor marking

Drop with P=1

Thmin ThmaxQavg

Pmax

0

Pdrop/mark

1

Average Queue Length Drop Probability P

Page 8: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

8

Drawbacks of Queue-based AQM

• Insensitive to current queue arrival and drain rates.• Long term queue length average produces slow

response.• Difficult to configure parameters.

Small drop probability From small queue

After some period

After some period

Large queue

Small queueLarge drop probability From large queue

Page 9: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

9

Objectives

• Is queue length a sufficient congestion indicator?

• Can we use load information for more precise congestion indication?

• Is queue length information still important, even when the load information is used?

• How do RED, SRED, FRED, BLUE, and REM compare in terms of user-perceivable metrics, such as Web response time?

• Can we achieve both high responsiveness and high throughput?

Page 10: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

10

Load/Delay Control (LDC)

• Load factor• R = queue arrival rate / queue service rate

• Provides better load (input/output rate) response and more intuitive parameters.

• Maintains RED benefits: misbehaving flow punishment and global synchronization avoidance.

• Unlike many load-based schemes, ECN is not required.

Page 11: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

11

Load/Delay Control (LDC)

• Uses both queue length information and load factor as multiple time scale congestion indicators.• Long-term: Queue length gives a more

stable (slower) congestion indication.• Short-term: Load factor enables faster

response.

• Goal: Maintain queuing delay below a target value and queue arrival rate below queue service rate.

Page 12: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

12

Load/Delay Control (LDC)

• Packet drop probability of LDC

))(,1min(4

1))(,1min(

4

3

target

avg

target

avgdrop

nn

R

R

Q

QP

Page 13: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

13

Performance Evaluation

• Network simulator ns-2.1b6• GFC-2 Configuration

• 22 HTTP on each sender-receiver pair, 1 unlimited FTP (H), 12 kbps CBR-UDP (F)

• Average 5 simulation runs, each runs 900 seconds

• Performance Metrics• Web response time (seconds), Goodput (Mbps),

Packet Drop ratio (%), Delay for UDP connections (seconds)

Page 14: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

14

Simulation Parameters

• RED• Thmin=buffer/12, Thmax=buffer/4, w_q=0.002, max_p=0.1,

gentle RED• SRED

• Full SRED, max_p=0.15, M=1000, p=0.25• FRED

• Min_q=4, others same as RED• BLUE

• Decrement =0.02, increment=0.2, freeze_time=0.1 s• REM

=0.1, =0.005, =1.001, w_q=0.002, =0.002 s• LDC

• D_target=4.0 s, (w_q=0.002, w_r=0.998, n=3, R_target=0.95, =0.002 s)

Page 15: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

15

Simulation Results

Algo-rithm

Web mean

Response

UDP Pkt

Delay

Web mean

Goodput

UDP mean

Goodput

FTP mean

Goodput

Pkt DropRatio

DropTail 79.07 0.95 0.110 0.12 0.134 1.31

RED 69.35 0.46 0.098 0.12 0.134 2.86

SRED 74.18 1.30 0.100 0.12 0.129 1.75

FRED 52.86 0.30 0.132 0.11 0.127 4.28

BLUE 79.02 1.40 0.096 0.12 0.132 1.47

REM 62.34 0.40 0.093 0.12 0.125 3.28

LDC 58.85 0.22 0.088 0.12 0.129 4.27

Page 16: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

16

Merits and Drawbacks of AQM

• RED• Merits

• Early congestion detection

• No bias against bursty traffic

• No global synchronization

• Drawbacks• Difficulty in

parameter setting• Insensitivity to traffic

load and drain rates

• SRED• Merits

• Stabilized queue occupancy

• Protection from misbehaving flows

• Drawbacks• Some per-flow state

(zombie list)• RED disadvantages

Page 17: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

17

Merits and Drawbacks of AQM

• FRED• Merits

• Good protection from misbehaving flows

• Drawbacks• Per-flow state• RED disadvantages

• BLUE• Merits

• Simplicity• High throughput

• Drawbacks• No early congestion

detection (Pdrop updated only on queue overflow or link idle events)

• Slow response and dependence on history

Page 18: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

18

Merits and Drawbacks of AQM

• REM• Merits

• Low delay and small queues

• Independence of the number of users

• Drawbacks• Some complexity due

to parameters• Low throughput for

Web traffic• Inconsistency with

TCP sender mechanism; works best with ECN

• LDC• Merits

• Sensitivity to traffic load and drain rate

• Low delay• Target delay achieved• Intuitive parameters,

meaningful to users (target delay)

• Drawbacks• Some complexity due

to parameters• Low throughput in

some cases

Page 19: A Comparison of  Load-based and Queue-based  Active Queue Management Algorithms

19

Conclusions and Future Work

• Queue-based algorithms are difficult to configure.

• Load-based algorithms reduce delay, but sometimes exhibit low throughput.

• A configurable combination that takes into consideration input rate, drain rate and queuing delay works well.

• Need to study LDC with ECN, parameter values of LDC, and the effect of variable output rate in multiple queues.