Transcript Active Queue Management Rong Pan Cisco System EE384y

Active Queue Management
Rong Pan
Cisco System
EE384y
Spring Quarter 2008
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
2
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
3
TCP+Queue Management
- A closed-loop control system
C
W/R
_
+

N
+

q
_
0.5
-
+
Time
Delay
p
Queue
Management
1
4
Drop Tail
- problems
• Lock out
• Full queue
• Bias against bursty traffic
• Global synchronization
5
Tail Drop Queue Management
Lock-Out
Max Queue Length
6
Tail Drop Queue Management
Full-Queue
• Only drop packets when queue is full
– long steady-state delay
7
Bias Against Bursty Traffic
Max Queue Length
8
Tail Drop Queue Management
Global Synchronization
Max Queue Length
9
Alternative Queue
Management Schemes
• Drop from front on full queue
• Drop at random on full queue
 both solve the lock-out problem
 both have the full-queues problem
10
Active Queue Management
Goals
• Solve tail-drop problems
–
–
–
–
no lock-out behavior
no full queue
no bias against bursty flow
no global synchronization
• Provide better QoS at network nodes
– low steady-state delay
– lower packet dropping
11
Random Early Detection
(RED)
Arriving packet
no
AvgQsize > Minth?
yes
Admit the
new packet
end
no
Admit packet with
a probability p
end
AvgQsize > Maxth?
yes
Drop the new packet
end
12
Drop Probability
RED Dropping Curve
1
maxp
0
minth
maxth
Average Queue Size
13
Effectiveness of RED
- Lock-Out & Global Synchronization
• Packets are randomly dropped
• Each flow has the same probability of being
discarded
14
Effectiveness of RED
- Full-Queue & Bias against bursty traffic
• Drop packets probabilistically in anticipation
of congestion
– not when queue is full
• Use qavg to decide packet dropping
probability: allow instantaneous bursts
15
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
16
Bad News - unresponsive end hosts
udp
udp
tcp
tcp
udp
tcp
tcp
The Internet
Connectionless; Best-Effort
17
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 with fairness
enhancement:
e.g. CHOKe, AFD, WRED
18
Active Queue Manament With
Enhancement to Fairness
FIFO
• Approximate fair bandwidth allocation
• Provide isolation from unresponsive flows
• Be as simple as RED
19
CHOKe
RED
Arriving packet
AvgQsize > Minth?
no
Drawyes
a packet at
random from queue
Admit the
new packet
end
no
yes
Flow id same as AvgQsize > Maxth?
nopacket id ?
the new
no
yes
Drop
packet
Admit packet with AvgQsize
Drop both
> the
Maxnew
th?
no
p
matched packets a probability
yes
end
end with
Admit packet
a probability p
end
end
Drop the new packet
end
20
Random Sampling from
Queue
UDP flow
•
A randomly chosen packet more likely
from the unresponsive flow
•
Adversary can’t fool the system
21
Comparison of Flow ID
•
Compare the flow id with the incoming packet
– more acurate
– Reduce the chance of dropping packets from a TCPfriendly flows.
22
Dropping Mechanism
• Drop packets (both incoming and matching
samples )
– More arrival -> More Drop
– Give users a disincentive to send more
23
Simulation Setup
S(1)
m TCP
Sources
S(2)
D(1)
10Mbps
10Mbps
D(2)
m TCP
Sinks
S(m)
S(m+1)
1Mbps
D(m)
D(m+1)
n UDP
Sources
n UDP
Sinks
S(m+n)
D(m+n)
24
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
25
32 TCP, 1 UDP (one sample)
Throughput (Kbps)
1000
800
UDP Throughput (RED)
UDP Throughput (CHOKe)
Avg. TCP Throughput (CHOKe)
600
400
74.1%
200 23.0%
0
100
99.6%
1000
UDP Arrival Rate (Kbps)
10000
26
32 TCP, 5 UDP (5 samples)
Throughput (Kbps)
1500
CHOKe(one sample):Total UDP Throughput
CHOKe(one sample):Total TCP Throughput
CHOKe with 5 samples: Total UDP Throughput
CHOKe with 5 samples: Total TCP Throughput
1200
900
600
300
0
100
1000
Total UDP Arrival Rate (Kbps)
10000
27
How Many Samples to
Take?
Rk
Maxth

R2 R1
avg
minth
Different samples for different Qlenavg
–
# samples  when Qlenavg close to minth
–
# samples when Qlenavg close to maxth
28
Throughput (Kbps)
32 TCP, 5 UDP (selfadjusting)
Total UDP Throughput
1200
Total TCP Throughput
900
600
61.1%
38.3%
300
0
89.7%
6.6%
100
99.3%
1000
10000
Total UDP Arrival Rate (Kbps)
29
Analytical Model
discards from the queue
permeable tube
with leakage
30
Fluid Analysis

N: the total number of packets in the buffer

Li(t): the survival 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 )
31
Model vs Simulation
- multiple TCPs and one UDP
350
1/(1+e)
Throughput
300
250
200
150
100
fluid model
CHOKe ns simulation
50
0
0.1
1
10
Arrival Rate
32
Fluid Model
- Multiple samples

Multiple samples are chosen
Li(t)t - Li(t +t)t = Mi t Li(t)t /N
- dLi(t)/dt = Mi Li(t) N
Li(0) = i (1-pi )M
Li(D) = i (1-pi )M - Mi pi
33
Two Samples
- multiple TCPs and one UDP
UDP Throughput(Kbps)
140
120
100
80
60
40
NS Simulation
Fluid Model
20
0
0
0.5
1
1.5
UDP Arrival Rate(Mbps)
2
34
Two Samples
- multiple TCPs and two UDP
UDP Throughput(Kbps)
200
160
120
80
NS Simulation
Fluid Model
40
0
0
0.5
1
1.5
2
UDP Arrival Rate(Mbps)
35
What If We Use a
Small Amount of
State?
36
AFD: Goal
• Approximate weighted bandwidth
allocation
– Not only AQM, approximate WDRR
scheduling
– Provide soft queues in addition to physical
queues
• Keep the state requirement small
• Be simple to implement
37
AFD Algorithm: Details
(Basic Case: Equal Share)
Di = Drop Probability for Class i
Arriving Packets
Qlen
1-Di
Qref
Class i
Di
Mi = Arrival estimate
for Class i
(Bytes over interval Ts)
Mfair = f(Qlen, Qlen_old,Qref)
Fair Share
If Mi  Mfair : No Drop (Di = 0)
If Mi > Mfair : Di > 0 such that
Mi (1-Di) = Mfair
38
AFD Algorithm: Details
(General Case)
Di = Drop Probability for Class i
Arriving Packets
Qlen
1-Di
Qref
Class i
Di
Mi = Arrival estimate
for Class i
(Bytes over interval Ts)
Mfair = f(Qlen, Qlen_old,Qref)
Fair Share
If Mi  F(Mfair,Mini,Maxi,Wi, …): No Drop (Di = 0)
If Mi > Mfair : Di > 0 such that
Mi (1-Di) = F(Mfair,Mini,Maxi,Wi, …)
39
Not Per-Flow State
Fraction of flows
• State requirement on the order of # of unresponsive flows
• Elephant Traps (developed jointly Stanford and Cisco)
40
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 each class)
• observation of queue length as measure of
congestion
– perform selective dropping (pre-enqueue) to drive
arrival rate to the desired departure rate
41
Mixed Traffic
with Different Levels of Unresponsiveness
Throughput (kbps)
500
400
300
200
100
0
RED
CHOKe
AFD
42
Drop Probabilities
(note differential dropping)
Drop Probability
0.3
0.25
0.2
0.15
0.1
0.05
0
RED
CHOKe
AFD
43
Different Number of TCP
Flows in Each Class
Class 2
Class 1
10 TCP Flows
5 TCP Flows
0
50
100
150
200
0
time
50
100
150
200
time
200
time
20 TCP Flows
Class 3
Class 4
15 TCP Flows
0
50
100
150
200
time
0
50
100
150
44
Different Class Throughput
Comparison
45
Queue Length
46
Mfair
47
AFD Implementation Issues
• Monitor Arrival Rate
• Determine Drop Probability
• Maximize Link Utilization
48
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
49
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)
50
Implementing the Drop
Function
• Drop levels are: 1/32, 1/16, 3/32…
• Suppose mi = 100, mfair = 62.0 => Di = 0.380,
Di
0.0
0.375 0.406
We choose the higher
value using binary
search
1.0
51
AFD - Summary
Ideal
FQ
Fairness
AFD
CHOKe
RED
Simplicity
• Equal share is approximated in a wide variety of settings
• The state requirement is limited
52
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
53