Active Queue Management

Rong PanCisco System

EE384ySpring Quarter 2006

2

Outline• Queue Management

– Drop as a way to feedback to TCP sources– Part of a closed-loop

• Traditional Queue Management – Drop Tail– Problems

• Active Queue Management– RED– CHOKe– AFD

3

Queue Management: Drops/Marks- A Feedback Mechanism To Regulate End TCP Hosts

• End hosts send TCP traffic -> Queue size

• Network elements, switches/routers, generate drops/marks based on their queue sizes

• Drops/Marks: regulation messages to end hosts

• TCP sources respond to drops/marks by cutting down their windows, i.e. sending rate

4

TCP+Queue Management- A closed-loop control system

QueueManagement

_

N

C

_

q

Time Delay

+

+

-

1

W/R

p

0.5

+

5

Drop Tail- problems

• Lock out

• Full queue

• Bias against bursty traffic

• Global synchronization

6

Max Queue Length

Tail Drop Queue ManagementLock-Out

7

Tail Drop Queue Management Full-Queue

• Only drop packets when queue is full– long steady-state delay

8

Max Queue Length

Bias Against Bursty Traffic

9

Max Queue Length

Tail Drop Queue ManagementGlobal Synchronization

10

• Drop from front on full queue

• Drop at random on full queue

both solve the lock-out problem both have the full-queues problem

Alternative Queue Management Schemes

11

• Solve tail-drop problems– no lock-out behavior– no global synchronization– no bias against bursty flow

• Provide better QoS at a router– low steady-state delay– lower packet dropping

Active Queue ManagementGoals

12

Random Early Detection (RED)

yes

Drop the new packet

end

Admit packet with a probability p

end

AvgQsize > Maxth?

yes

Arriving packet

no

Admit the new packet

end

AvgQsize > Minth? no

13

RED Dropping Curve

minth maxth

0

Average Queue Size

Dro

p P

roba

bilit

y 1

maxp

14

Effectiveness of RED- Lock-Out & Global Synchronization

• Packets are randomly dropped

• Each flow has the same probability of being discarded

15

• Drop packets probabilistically in anticipation of congestion – not when queue is full

• Use qavg to decide packet dropping probability: allow instantaneous bursts

Effectiveness of RED- Full-Queue & Bias against bursty traffic

16

What QoS does RED Provide?

• Lower buffer delay: good interactive service

– qavg is controlled to be small

• Given responsive flows: packet dropping is reduced– early congestion indication allows traffic to throttle back before congestion

• Given responsive flows: fair bandwidth allocation

17

Bad News - unresponsive end hosts

tcp

tcpudp

udp

udp

tcp

tcp

Connectionless; Best-Effort

The Internet

18

Scheduling & Queue Management

• What routers want to do?– isolate unresponsive flows (e.g. UDP)

– provide Quality of Service to all users

• Two ways to do it– scheduling algorithms:

e.g. FQ, CSFQ, SFQ

– queue management algorithms:

e.g. RED, FRED, SRED

19

FQ vs. RED

• Hard/Expensive to implement

• Isolation from non-adaptive flows

FQ

RED

• No isolation from non-adaptive flows

• Easy to implement

20

Active Queue Manament With Enhancement to Fairness

• Provide isolation from unresponsive flows• Be as simple as RED

FIFO

21

yes

Drop the new packet

end

Admit packet with a probability p

end

AvgQsize > Maxth?

yes

REDArriving packet

no

Admit the new packet

end

AvgQsize > Minth? no

yes

no

Drop both matched packets

end

Draw a packet atrandom from queue

Flow id same as the new packet id ?

yes

Drop the new packet

end

Admit packet with a probability p

end

noAvgQsize > Maxth?

no

CHOKe

22

Random Sampling from Queue

• A randomly chosen packet more likely from the

unresponsive flow

• Adversary can’t fool the system

UDP flow

23

Comparison of Flow ID

• Compare the flow id with the incoming packet

– more acurate

– Reduce the chance of dropping packets from a TCP-friendly flows.

24

Dropping Mechanism

• Drop packets (both incoming and matching samples )

– More arrival -> More Drop

– Give users a disincentive to send more

25

1Mbps

10MbpsS(2)

S(m)

S(m+n)

m TCP Sources

S(m+1)

n UDP Sources

D(2)

D(m)

D(m+n)

m TCP Sinks

D(m+1)

n UDP Sinks

D(1)

10Mbps

Simulation SetupS(1)

26

Network Setup Parameters 32 TCP flows, 1 UDP flow All TCP’s maximum window size = 300 All links have a propagation delay of 1ms FIFO buffer size = 300 packets All packets sizes = 1 KByte RED: (minth,maxth) = (100,200) packets

27

32 TCP, 1 UDP (one sample)

0

200

400

600

800

1000

100 1000 10000UDP Arrival Rate (Kbps)

Thro

ughp

ut (K

bps)

UDP Throughput (RED)UDP Throughput (CHOKe)Avg. TCP Throughput (CHOKe)

23.0%

74.1%

99.6%

28

32 TCP, 5 UDP (5 samples)

0

300

600

900

1200

1500

100 1000 10000Total UDP Arrival Rate (Kbps)

Thro

ughp

ut (K

bps)

CHOKe(one sample):Total UDP ThroughputCHOKe(one sample):Total TCP ThroughputCHOKe with 5 samples: Total UDP ThroughputCHOKe with 5 samples: Total TCP Throughput

29

How Many Samples to Take?

minthMaxth

R1R2Rk

Different samples for different Qlenavg

– # samples when Qlenavg close to minth

– # samples when Qlenavg close to maxth

avg

30

6.6%

38.3%61.1%

89.7%

99.3%0

300

600

900

1200

100 1000 10000

Total UDP Arrival Rate (Kbps)

Th

rou

gh

pu

t (K

bp

s)

Total UDP Throughput

Total TCP Throughput

32 TCP, 5 UDP (self-adjusting)

31

Analytical Model

discards from the queue

permeable tube with leakage

32

Fluid Analysis

N: the total number of packets in the buffer Li(t): the survial rate for flow i packets

Li(t)t - Li(t +t)t = i t Li(t)t /N

- dLi(t)/dt = i Li(t) N

Li(0) = i (1-pi )

Li(D) = i (1-2pi )

33

Model vs Simulation- multiple TCPs and one UDP

0

50

100

150

200

250

300

350

0.1 1 10Arrival Rate

Thr

ough

put

fluid model

CHOKe ns simulation

1/(1+e)

34

Fluid Model - Multiple samples

Li(t)t - Li(t +t)t = Mi t Li(t)t /N

- dLi(t)/dt = Mi Li(t) N

Li(0) = i (1-pi )M

Li(D) = i (1-pi )M - Mi pi

Multiple samples are chosen

35

Two Samples- multiple TCPs and one UDP

0

20

40

60

80

100

120

140

0 0.5 1 1.5 2UDP Arrival Rate(Mbps)

UD

P T

hrou

ghpu

t(K

bps)

NS SimulationFluid Model

36

Two Samples- multiple TCPs and two UDP

0

40

80

120

160

200

0 0.5 1 1.5 2

UDP Arrival Rate(Mbps)

UD

P T

hrou

ghpu

t(K

bps)

NS Simulation

Fluid Model

37

What If We Use a Small Amount of

State?

38

AFD: Goal

• Approximate equal bandwidth allocation– Not only AQM, approximate DRR scheduling

– Provide soft queues in addition to physical queues

• Keep the state requirement small

• Be simple to implement

39

AFD Algorithm: Details (Basic Case: Equal Share)

Di = Drop Probability for Class i

Qref

1-Di

Di

QlenArriving Packets

If Mi > Mfair : Di > 0 such that Mi (1-Di) = Mfair

If Mi Mfair : No Drop (Di = 0)

Mi = Arrival estimatefor Class i

(Bytes over interval Ts)

Mfair = Mfair - a (Qlen - Qref)+ b (Qlen_old - Qref)

Fair Share

Class i

40

AFD Algorithm: Details (General Case)

Di = Drop Probability for Class i

Qref

1-Di

Di

QlenArriving Packets

If Mi > Mfair : Di > 0 such that Mi (1-Di) = F(Mfair,Mini,Maxi,Wi, …)

If Mi F(Mfair,Mini,Maxi,Wi, …): No Drop (Di = 0)

Mi = Arrival estimatefor Class i

(Bytes over interval Ts)

Mfair = Mfair - a (Qlen - Qref)+ b (Qlen_old - Qref)

Fair Share

Class i

41

Not Per-Flow State

Fraction of flows

• State requirement on the order of # of unresponsive flows

42

AFD Solution: Details

• Based on 3 simple mechanisms– estimate per “class” arrival rate

• counting per “class” bytes over fixed intervals ( Ts )• potential averaging over multiple intervals

– estimate deserved departure rate (so as to achieve the proper bandwidth allocation for the class)

• Observation and averaging of queue length as measure of congestion

• Functional definition of “fair share” based on fairness criterion

– perform probabilistic dropping (pre-enqueue) to drive arrival rate to equal desired departure rate

43

Mixed Trafficwith Different Levels of Unresponsiveness

0

100

200

300

400

500

RED CHOKe AFD

Thr

ough

put (

kbps

)

44

Drop Probabilities(note differential dropping)

0

0.05

0.1

0.15

0.2

0.25

0.3

RED CHOKe AFD

Dro

p P

roba

bili

ty

45

Different Number of TCP Flows in Each Class

Cla

ss 1

5 TCP Flows

0 50 150 200100 time

Cla

ss 2

10 TCP Flows

0 50 150 200100 time

Cla

ss 3

15 TCP Flows

0 50 150 200100 time

Cla

ss 4

20 TCP Flows

0 50 150 200100 time

46

Different Class Throughput Comparison

47

Queue Length

48

Mfair

49

AFD Implementation Issues

• Monitor Arrival Rate

• Determine Drop Probability

• Maximize Link Utilization

50

Arrival Monitoring

• Keep a counter for each class– Count the data arrivals (in bytes) of each

class in 10ms interval: arvi

• Arrival rate of each class is updated every 10ms– mi = mi(1-1/2c)+arvi

– c determines the average window

51

Implementing the Drop Function

• If Mi Mfair then Di = 0

• Otherwise, rewrite the drop function as

• Suppose we have predetermined drop levels, find the one such

that Di* Mi = (Mi – Mfair)

52

Implementing the Drop Function

• Suppose mi = 100, mfair = 62.0 => Di = 0.380,

0.0 1.0

Di

We choose the higher value using

binary search

0.375 0.406

• Drop levels are: 1/32, 1/16, 3/32…

53

FQ

RED

Simplicity

Fai

rnes

s

AFD - Summary

• Equal share is approximated in a wide variety of settings

• The state requirement is limited

CHOKeAFD

Ideal

54

Summary

• Traditional Queue Management – Drop Tail, Drop Front, Drop Random– Problems: lock-out, full queue, global

synchronization, bias against bursty traffic

• Active Queue Management– RED: can’t handle unresponsive flows– CHOKe: penalize unresponsive flows– AFD: provides approximate fairness with

limited states