csit560 internet infrastructure: switches and routers
Post on 31-Jan-2016
32 Views
Preview:
DESCRIPTION
TRANSCRIPT
CSIT560CSIT560Internet Infrastructure:Internet Infrastructure:Switches and RoutersSwitches and Routers
Active Queue ManagementActive Queue Management
Presented By: Gary Po, Henry Hui and Kenny Chong
AgendaAgenda
Why AQM?Why AQM? In ConsiderationsIn Considerations AQM AlgorithmsAQM Algorithms Commercial EffortCommercial Effort ConclusionsConclusions
Why AQM?Why AQM?
Two Classes of Router Algorithms for Two Classes of Router Algorithms for Congestion ControlCongestion Control
What is Congestion?What is Congestion? Congestion avoidance in TCP, is it Congestion avoidance in TCP, is it
good enough?good enough? Our Choice :Our Choice :
Active Queue ManagementActive Queue Management
What do we consider when What do we consider when implementing AQM?implementing AQM?
QoSQoS– Keep Average queue size smallKeep Average queue size small– Bounded DelayBounded Delay
Link UtilizationLink Utilization– Avoid Global SynchronizationAvoid Global Synchronization– Absorbs bursts without dropping packetsAbsorbs bursts without dropping packets
FairnessFairness– Punishes misbehaving flowsPunishes misbehaving flows– Prevent bias against bursty connectionsPrevent bias against bursty connections
ImplementationImplementation– Ease of ConfigurationsEase of Configurations– Buffer Size Requirement (Large or Small)Buffer Size Requirement (Large or Small)– Per-flow State InformationPer-flow State Information– Computational OverheadComputational Overhead
AQM AlgorithmsAQM Algorithms
FIFO + Drop TailFIFO + Drop Tail
Problems:Problems:– No isolationNo isolation– No policingNo policing– Large queues for high utilizationsLarge queues for high utilizations– Synchronization problemSynchronization problem– Lock-out problemLock-out problem
Define Two Threshold Values
RED (Random Early Detection)RED (Random Early Detection)
FIFO schedulingFIFO scheduling
Min threshMax thresh
Average queuelength
Make Use of Average Queue LengthCase 1:
Average Queue Length < Min. Thresh ValueAdmit the New Packet
RED (Cont’d)RED (Cont’d)
Min threshMax thresh
Average queuelength
Case 2: Average Queue Length betweenMin. and Max. Threshold Value
p
1-p
Admit the New Packet With Probability p…
p
1-p
Or Drop the New Packet With Probability 1-p
RED (Cont’d)RED (Cont’d)
Min threshMax thresh
Average queuelength
Case 3: Avg. Queue Length > Max. Threshold Value
New Packet will be droppedAs no new packets can be admitted,the average queue length decreases.
Until the average queue length drops belowthe max threshold value
New packet could be admitted with a probability p…
or being dropped with a probability 1-p…
RED Flow DiagramRED Flow Diagram
RED (Cont’d)RED (Cont’d)
Queue Size versus TimeQueue Size versus Time
0 5 10 15 20 25 30 35 40 45 500
50
100
150
200
250CHOKe: Queue Size
Time (Seconds)
Siz
e of
Que
ue (
No.
of
Pac
kets
)
Average Queue SizeCurrent Queue Size
Delay is bounded
Delay is bounded
Global Synchronization solvedGlobal Synchronization solved
RED: Queue Size
Unfairness of REDUnfairness of RED
0 5 10 15 20 25 30 350
100
200
300
400
500
600
700
800
900
1000RED's Throughput
Flow Number
Thr
ough
put
(Kbp
s)Idea Fair ShareRED's Throughput
Unresponsive Flow (such
as UDP)
Unresponsive Flow (such
as UDP)
32 TCP Flows1 UDP Flow
32 TCP Flows1 UDP Flow
An unresponsiveflow occupies over 95% of bandwidth
An unresponsiveflow occupies over 95% of bandwidth
CHOKeCHOKe(CHOose and Keep)(CHOose and Keep)
Based on REDBased on RED SimpleSimple Designed for fairnessDesigned for fairness Penalize the unresponsive flowPenalize the unresponsive flow
CHOKe (Cont’d)CHOKe (Cont’d)
MechanismMechanism
CHOKe (Cont’d)CHOKe (Cont’d)
Min threshMax thresh
Average queuelength
Case 1: Average Queue Length < Min. Thresh Value
Admit the New Packet
CHOKe (Cont’d)CHOKe (Cont’d)
Min threshMax thresh
Average queuelength
p
1-p
Case 2: Avg. Queue Length is between Min. and Max. Threshold Values
A packet is randomly chosen from the queue to compare with the new arrival packet
If they are from different flows, the samelogic in RED applies
If they are from the same flow, both packets will be dropped
CHOKe (Cont’d)CHOKe (Cont’d)
Min threshMax thresh
Average queuelength
Case 3:Avg. Queue Length > Max. Threshold Value
A random packet will be chosen forcomparison
If they are from different flows, the new packet will be droppedIf they are from the same flow,
both packets will be dropped
Evaluate CHOKe’s Evaluate CHOKe’s performance using NS-2performance using NS-2
Simulation ScenarioSimulation Scenario
10Mbps
1Mbps
10Mbps
UDP
TCP
UDP
TCP
source destination
router router
• Topology: Dumb-bell• Metrics: throughput and queue size
Performance of CHOKePerformance of CHOKe
0 5 10 15 20 25 30 350
5
10
15
20
25
30
35
40
Flow Number
Thr
ough
put
(Kbp
s)
CHOKe: Throughput Per Flow
Ideal Fair ShareCHOKe
Fair Share Level
Fair Share Level
Bandwidth is evenly shared
Bandwidth is evenly shared
Unresponsive Flow (UDP)Unresponsive Flow (UDP)
32 TCP Flows1 UDP Flow
32 TCP Flows1 UDP Flow
ParametersParameters
Number of responsive/unresponsive Number of responsive/unresponsive flowsflows
Transfer rate of different flowsTransfer rate of different flows Number of random candidates Number of random candidates
chosen for comparisonchosen for comparison
CHOKe SimulationCHOKe Simulation
Different Parameters, different performanceDifferent Parameters, different performance
0 5 10 15 20 25 30 350
50
100
150
200
250CHOKe 1: Throughput Per Flow
Flow Number
Thr
ough
put
(Kbp
s)
Ideal Fair ShareCHOKe 1
0 5 10 15 20 25 30 350
50
100
150CHOKE 2: Throughput with 3 UDP with Different Rate
Flow Number
Thr
ough
put
Ideal Fair ShareCHOKE 2
0 5 10 15 20 25 30 350
5
10
15
20
25
30
35
40
45
50CHOKe 2: Throughput Per Flow (with 4x UDP rate)
Flow Number
Thr
ough
put
(Kbp
s)
Ideal Fair ShareCHOKe 2
0 5 10 15 20 25 30 350
50
100
150
200
250CHOKe 2: Throuhput Per Flow (with 3 UDP Flows)
Flow Number
Thr
ough
put
(Kbp
s)
Ideal Fair ShareCHOKe 2
CHOKe-132 TCPs1 UDP
CHOKe-132 TCPs1 UDP
CHOKe-232 TCPs,1 UDP of high rate
CHOKe-232 TCPs,1 UDP of high rate
CHOKe-232 TCPs,3 UDPs of
different rate
CHOKe-232 TCPs,3 UDPs of
different rate
CHOKe-232 TCPs,3 UDPs of same rate
CHOKe-232 TCPs,3 UDPs of same rate
Evolutions of AQM AlgorithmsEvolutions of AQM Algorithms
FIFO+FIFO+DropTailDropTail
REDRED
FREDFRED
CHOKeCHOKe SACSAC
BLUEBLUE SFBSFB
SREDSRED
REM, AVQ, PI ControllerREM, AVQ, PI Controller REDRED– MeritsMerits
Early congestion Early congestion detectiondetection
No bias against No bias against bursty trafficbursty traffic
No global No global synchronizationsynchronization
– DrawbacksDrawbacks Difficulty in Difficulty in
parameter settingparameter setting Insensitivity to traffic Insensitivity to traffic
load and drain ratesload and drain rates
REDRED– MeritsMerits
Early congestion Early congestion detectiondetection
No bias against No bias against bursty trafficbursty traffic
No global No global synchronizationsynchronization
– DrawbacksDrawbacks Difficulty in Difficulty in
parameter settingparameter setting Insensitivity to traffic Insensitivity to traffic
load and drain ratesload and drain rates
SREDSRED– MeritsMerits
Stabilized queue Stabilized queue occupancyoccupancy
Protection from Protection from misbehaving flowsmisbehaving flows
– DrawbacksDrawbacks Some per-flow state Some per-flow state
(zombie list)(zombie list) RED disadvantagesRED disadvantages
SREDSRED– MeritsMerits
Stabilized queue Stabilized queue occupancyoccupancy
Protection from Protection from misbehaving flowsmisbehaving flows
– DrawbacksDrawbacks Some per-flow state Some per-flow state
(zombie list)(zombie list) RED disadvantagesRED disadvantages
FREDFRED– MeritsMerits
Good protection from Good protection from misbehaving flowsmisbehaving flows
– DrawbacksDrawbacks Per-flow statePer-flow state RED disadvantagesRED disadvantages
FREDFRED– MeritsMerits
Good protection from Good protection from misbehaving flowsmisbehaving flows
– DrawbacksDrawbacks Per-flow statePer-flow state RED disadvantagesRED disadvantages
BLUEBLUE– MeritsMerits
SimplicitySimplicity High throughputHigh throughput
– DrawbacksDrawbacks No early congestion No early congestion
detection (Pdrop detection (Pdrop updated only on updated only on queue overflow or queue overflow or link idle events)link idle events)
Slow response and Slow response and dependence on dependence on historyhistory
BLUEBLUE– MeritsMerits
SimplicitySimplicity High throughputHigh throughput
– DrawbacksDrawbacks No early congestion No early congestion
detection (Pdrop detection (Pdrop updated only on updated only on queue overflow or queue overflow or link idle events)link idle events)
Slow response and Slow response and dependence on dependence on historyhistory
REMREM– MeritsMerits
Low delay and small Low delay and small queuesqueues
Independence of the Independence of the number of usersnumber of users
– DrawbacksDrawbacks Some complexity due to Some complexity due to
parametersparameters Low throughput for Web Low throughput for Web
traffictraffic Inconsistency with TCP Inconsistency with TCP
sender mechanism; sender mechanism; works best with ECNworks best with ECN
REMREM– MeritsMerits
Low delay and small Low delay and small queuesqueues
Independence of the Independence of the number of usersnumber of users
– DrawbacksDrawbacks Some complexity due to Some complexity due to
parametersparameters Low throughput for Web Low throughput for Web
traffictraffic Inconsistency with TCP Inconsistency with TCP
sender mechanism; sender mechanism; works best with ECNworks best with ECN
LDCLDC– MeritsMerits
Sensitivity to traffic load Sensitivity to traffic load and drain rateand drain rate
Low delayLow delay Target delay achievedTarget delay achieved Intuitive parameters, Intuitive parameters,
meaningful to users meaningful to users (target delay)(target delay)
– DrawbacksDrawbacks Some complexity due to Some complexity due to
parametersparameters Low throughput in some Low throughput in some
casescases
LDCLDC– MeritsMerits
Sensitivity to traffic load Sensitivity to traffic load and drain rateand drain rate
Low delayLow delay Target delay achievedTarget delay achieved Intuitive parameters, Intuitive parameters,
meaningful to users meaningful to users (target delay)(target delay)
– DrawbacksDrawbacks Some complexity due to Some complexity due to
parametersparameters Low throughput in some Low throughput in some
casescases
Commercial Efforts & Commercial Efforts & ConclusionConclusion
Commercial Efforts & Commercial Efforts & Conclusion (Cont’d)Conclusion (Cont’d)
““ Applying AQM over 3G wireless Applying AQM over 3G wireless network” – a paper supported by network” – a paper supported by Motorola Canada Ltd. (Mar. 2003)Motorola Canada Ltd. (Mar. 2003)
3G network, real-time applications 3G network, real-time applications have hard time deadlines for packet have hard time deadlines for packet delivery at the receiver.delivery at the receiver.
Use AQM to avoid long queuing delay Use AQM to avoid long queuing delay and prevent expiring packets.and prevent expiring packets.
Commercial Efforts & Commercial Efforts & Conclusion (Cont’d)Conclusion (Cont’d)
AQM improves overall system AQM improves overall system performance by increasing performance by increasing throughput and reducing end-to-end throughput and reducing end-to-end delay.delay.
Commercial Efforts & Commercial Efforts & Conclusion (Cont’d)Conclusion (Cont’d)
““ Effect of AQM on Web Effect of AQM on Web Performance” – a paper supported by Performance” – a paper supported by Cisco Systems and IBM. (Aug. 2003)Cisco Systems and IBM. (Aug. 2003)
Proportional Integrator (PI) controllerProportional Integrator (PI) controller Random Exponential Marking (REM)
controller Adaptive Random Early Detection (ARED). IETF proposed standard :
Explicit Congestion Notification (ECN)
Commercial Efforts & Commercial Efforts & Conclusion (Cont’d)Conclusion (Cont’d)
ECN has significant impact with AQM ECN has significant impact with AQM scheme in web performance.scheme in web performance.
Many researches and efforts are Many researches and efforts are going on in the field of AQM.going on in the field of AQM.
Simple and Easy to implement Simple and Easy to implement
top related