Performance & QoS CongestionControl · Congestion avoidance Packet marking Traffic shaping Traffic...

Performance&QoSCongestion Control

ETSF05/ETSF10– InternetProtocols

QualityofService(QoS)

• Maintainingafunctioningnetwork–Meetingapplications’demands

• User’sdemands=QoE (QualityofExperience)

– Dealingwithflowcharacteristics

3ETSF05/ETSF10- InternetProtocols

Jitter = Packet Delay Variations

Figure 22.1 Architectural Framework for QoS Support

Queuemanagement

Congestionavoidance

Packetmarking

Trafficshaping

Trafficpolicing

Trafficclassification

Queueing &scheduling

Data Plane

Admissioncontrol

QoSrouting

Resourcereservation

Control Plane

Servicelevel

agreement

Trafficrestoration

Policy

Trafficmetering

andrecording

Manag

emen

t

Plane

ETSF05/ETSF10- InternetProtocols 4

DataPlane• Includesthosemechanismsthatoperatedirectlyonflowsofdata– Queuemanagementalgorithms

• TaildropvsRED(RandomEarlyDetection)– Queueingandscheduling– Congestionavoidance– Packetmarking– Trafficclassification– Trafficpolicing– Trafficshaping


ControlPlane

• Concernedwithcreatingandmanagingthepathwaysthroughwhichuserdataflows– Admissioncontrol– QoS routing– Resourcereservation


ManagementPlane

• Containsmechanismsthataffectbothcontrolplaneanddataplanemechanisms– Servicelevelagreement(SLA)– Trafficmeteringandrecording– Trafficrestoration– Policy


Networkperformance• Datarate(Bandwidth)– Bitspersecond

• Throughput– Efficiency,alwayslessthancapacity(<1)– Alternatively:availabledatarate

• Latency(Delay)– Transmission,propagation,processing,queueing– OnewayorRTT(RoundTriptime)

• PDV=PacketDelay Variation(Jitter)– Real-time applications!


Otherparameters

• BitErrorRate– L1parameterthatheavilyimpactsonL3– Frame/PacketLossonhigherlayers

• InterPacketGapvariations– “Jitter”– Couldbenon-zeroalreadyatsender

• Ratioofpacketsoutoforder– ImpactondelayinTCP


Packetloss• Due to– Biterror inpacket

• Routersdiscard erronous packet• LinkorPhysical Layer?

– Queue overflow• Discard packets• Node problems

• Inreal-timemultimedialatepacketsconsidered lost• Packetlossratio (%)• NoteTCP’s sensitivity to packetloss


Circuit Switching Datagram Packet Switching Virtual Circuit Packet Switching

Dedicated transmission path No dedicated path No dedicated path

Continuous transmission of data

Transmission of packets Transmission of packets

Fast enough for interactive Fast enough for interactive Fast enough for interactive Messages are not stored Packets may be stored until

delivered Packets stored until delivered

The path is established for entire conversation

Route established for each packet

Route established for entire conversation

Call setup delay; negligible transmission delay

Packet transmission delay Call setup delay; packet transmission delay

Busy signal if called party busy

Sender may be notified if packet not delivered

Sender notified of connection denial

Overload may block call setup; no delay for established calls

Overload increases packet delay

Overload may block call setup; increases packet delay

Electromechanical or computerized switching nodes

Small switching nodes Small switching nodes

User responsible for message loss protection

Network may be responsible for individual packets

Network may be responsible for packet sequences

Usually no speed or code conversion

Speed and code conversion Speed and code conversion

Fixed bandwidth Dynamic use of bandwidth Dynamic use of bandwidth No overhead bits after call setup

Overhead bits in each packet Overhead bits in each packet

Table9.1

ComparisonofCommunication

SwitchingTechniques

(Tablecanbefoundonpage315in

textbook)


(a) Circuit switching

Callrequestsignal

Callacceptsignal

propagationdelay

processingdelay

(b) Virtual circuit packet switching

Callrequestpacket

Callacceptpacket

link link link

1 2 3 4Nodes: 1 2 3 4

Acknowledge-ment signal

Acknowledge-ment packet

(c) Datagram packet switching

1 2 3 4

Userdata

Pkt1

Pkt2

Pkt3Pkt1

Pkt2

Pkt3Pkt1

Pkt2

Pkt3

Pkt1

Pkt2

Pkt3Pkt1

Pkt2

Pkt3Pkt1

Pkt2

Pkt3

Figure 9.15 Event Timing for Circuit Switching and Packet SwitchingETSF05/ETSF10- InternetProtocols 12

VirtualCircuitsvs.Datagram• Virtualcircuits– Networkcanprovidesequencinganderrorcontrol– Packetsareforwardedmorequickly– Lessreliable(compareCircuitSwitching)

• Datagram(BestEffort)– Nocallsetupphase– Individualpackethandling–Moreflexible–Morereliable


IPPerformanceMetrics(IPPMwg)• CharteredbyIETFtodevelopstandardmetricsthatrelatetothequality,performance,andreliabilityofInternetdatadelivery

• Measurementtechniques– Active:Transmitpacketsovernetworkformeasurementpurposes

– Passive:Useexistingtrafficformeasurements


Table22.3IPPerformanceMetrics

Metric Name Singleton Definition Statistical Definitions One-Way Delay Delay = dT, where Src transmits first bit of

packet at T and Dst received last bit of packet at T + dT

Percentile, median, minimum, inverse percentile

Round-Trip Delay Delay = dT, where Src transmits first bit of packet at T and Src received last bit of packet immediately returned by Dst at T + dT

Percentile, median, minimum, inverse percentile

One-Way Loss Packet loss = 0 (signifying successful transmission and reception of packet); = 1 (signifying packet loss)

Average

One-Way Loss Pattern

Loss distance: Pattern showing the distance between successive packet losses in terms of the sequence of packets Loss period: Pattern showing the number of bursty losses (losses involving consecutive packets)

Number or rate of loss distances below a defined threshold, number of loss periods, pattern of period lengths, pattern of inter-loss period lengths.

Packet Delay Variation

Packet delay variation (pdv) for a pair of packets with a stream of packets = difference between the one-way-delay of the selected packets

Percentile, inverse percentile, jitter, peak-to-peak pdv

Src=IPaddressofahostDst =IPaddressofahost

(a)SampledmetricsETSF05/ETSF10- InternetProtocols 15

Table22.3IPPerformanceMetrics

Metric Name General Definition Metrics Connectivity Ability to deliver a packet

over a transport connection.

One-way instantaneous connectivity, Two-way instantaneous connectivity, one-way interval connectivity, two-way interval connectivity, two-way temporal connectivity

Bulk Transfer Capacity

Long-term average data rate (bps) over a single congestion-aware transport connection.

BTC = (data sent)/(elapsed time)

(b)Othermetrics


P(i) P(j)

P(j)

P(k)

P(k)P(i)

dTi

I1MP1

MP2

Figure 22.12 Model for Defining Packet Delay Variation

I1

I2

I2dTj dTk

I1, I2 = times that mark that beginning and ending of the interval in which the packet stream from which the singleton measurement is taken occurs.MP1, MP2 = source and destination measurement pointsP(i) = ith measured packet in a stream of packetsdTi = one-way delay for P(i)


PDVi=dTi-dTi-1Alt.STD(dT)

Time synch!

• How much datafills thelink• OneWayDelay• TwoWayDelay=RoundTripTime(RTT)

Timefordata+timeforACK


Bandwidth-delayproduct


bandwidth:5bps,delay:5sbandwidthxdelay=25bits



• Importantforcongestionavoidance– Don’toverfillthelink

• Importantforefficiency– Keepthelinkfilledatalltimes– FormaxefficiencyDatachunks>2*bandwidth*delay



• Important fortuning (TCP)• LongFatNetwork (LFN,”elephant”)BDP>>105 bits

• Very long(high delay)links:->Bandwidth =BDP/delayBut ittakes longtime before ACKarrives …


Performance vsARQ

• Method– Stop-&-Wait– Go-Back-N– Selective-Repeate

• Utilisation =f(window size)


t0 T RT R

t0 + 1 T RT R

t0 + a T RT R

t0 + 1 + a T RT R

t0 + 1 + 2a ACK

Frame

t0

t0 + a

t0 + 1

t0 + 1 + a

t0 + 1 + 2a T RT R

(b) a > 1(a) a < 1

Figure 16.8 Stop-and-Wait Link Utilization (transmission time = 1; propagation time = a)ETSF05/ETSF10- InternetProtocols 23

(a) W��a + 1

�a + 1

a + 1

a

�

1

t = 0

�)UDPH��a + 1)A BA�)UDPH��a) Frame (a�� Frame (a+3)

Frame (a + 1)A BA Frame a �)UDPH��

Frame aA B Frame (a – 1) Frame 1�)UDPH��

Frame 3

�)UDPH��A B Frame 1

Frame 1A B

A B

(b) W��a + 1

�a + 1

W

1

t = 0

A BA Frame (a ��

Frame WA BA Frame (W – 1) Frame (W–a+1)

Frame W

Frame (W-a��

a + 1 Frame (a + 1)A BA Frame a �)UDPH�� Frame 3

a Frame aA B Frame (a – 1) Frame 1�)UDPH��

Frame 1A B

A B

Figure 16.9 Timing of Sliding-Window Protocol

Sliding Windowsbased• a =propagationtime

• W =window size

• Compare withBandwitdh-DelayProduct


2015-12-07 ETSF05/ETSF10- InternetProtocols 25

(a) W��a + 1

�a + 1

a + 1

a

�

1

t = 0




Frame 3


Frame 1A B

A B

(b) W��a + 1

�a + 1

W

1

t = 0



Frame W

Frame (W-a��



Frame 1A B

A B


2015-12-07 ETSF05/ETSF10- InternetProtocols 26

(a) W��a + 1

�a + 1

a + 1

a

�

1

t = 0




Frame 3


Frame 1A B

A B

(b) W��a + 1

�a + 1

W

1

t = 0



Frame W

Frame (W-a��



Frame 1A B

A B


1.0

0.8

0.6

0.4

0.2

0.0

Util

izat

ion

0 .1 1 1 0 1 0 0 1000a

W = 7 W = 1

W = 127

Figure 16.10 Sliding-Window Utilization as a function of a


W=1 Stop-&-Wait

1.0

0.8

0.6

0.4

0.2

0.0

Util

izat

ion

0 .1 1 1 0 1 0 0 1000a

Stop-and-wait

W =127 Go-back-N

W =7 Go-back-N

Figure 16.11 ARQ Utilization as a Function of a (P = 10–3)

ETSF05/ETSF10- InternetProtocols 28P=Propability of frame error

1

Data1Data

Header

(a) 1-packet message (b) 2-packet message (c) 5-packet message (d) 10-packet message

Data

Data

Data2

Data1

Data2

Data1

Data2

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

1

2

3

4

5

6

7

8

9

10

2

3

4

5

1

2

3

4

5

1

2

3

4

5

Figure 9.13 Effect of Packet Size on Transmission Time

X a b Y

X a b Y

X a b Y

X a b Y

PacketSize vsTransmissionTime


Nor

mal

ized

Thr

ough

put

Load

Average delayfor packets thatare delivered

Average delayfor all packets

Del

ay

Load

Figure 20.4 The Effects of Congestion

No congestion Severe congestionModeratecongestion

A

B


Delay andthroughput:Finite buffersNocongestioncontrol

CongestionControlinPacket-SwitchingNetworks

Sendcontrolpackettosomeorallsourcenodes• Requiresadditionaltrafficduringcongestion

Relyonroutinginformation• Mayreacttooquickly

Endtoendprobepackets• Addstooverhead

Addcongestioninformationtopacketsintransit• Eitherbackwardsorforwards


ExplicitCongestionSignaling

• Backward– Congestionavoidancenotificationinoppositedirectiontopacketrequired

• Forward– Congestionavoidancenotificationinsamedirectionaspacketrequired


Congestion control

• Avoiding andeliminating congestion– Open-loop=proactive,prevent congestion– Closed-loop=reactive,control congestion


Closed-loopcongestion control (1)

• Backpressure


Closed-loopcongestion control (2)

• Chokepacket


ImplicitCongestionSignaling

• Withnetworkcongestion:– Transmissiondelayincreases– Packetsmaybediscarded(Packetloss)

• Sourcecandetectcongestionandreduceflow• Responsibilityofendsystems• Effectiveonconnectionless(datagram)networks

• Alsousedinconnection-orientednetworks


ExplicitSignalingCategories


• Binary– Abitsetinapacketindicatescongestion

• Creditbased– Indicateshowmanypacketssourcemaysend– Commonforend-to-endflowcontrol

• Ratebased– Supplyexplicitdataratelimit– Nodesalongpathmayrequestratereduction

How toimprove QoS?

• Admission control• Resource reservation• Scheduling• Traffic shaping

• Routing?


WheretoimproveQoS?

• Admissioncontrol– DIFFSERV:Serviceclasses– INTSERV:Reservationarchitectures

• Resource reservation– RSVP(Resource ResevationProtocol)

• Scheduling• Trafficshaping


ANYWHEREYOUFINDQUEUES!

Scheduling:FIFOqueuing


Scheduling:Priority queuing


Scheduling:Weightedfairqueuing


TrafficShaping/TrafficPolicing• Twoimportanttoolsinnetworkmanagement:– Trafficshaping

• Concernedwithtrafficleavingtheswitch• Reducespacketclumping• Producesanoutputpacketstreamthatislessburstandwithamoreregularflowofpackets

– Trafficpolicing• Concernedwithtrafficenteringtheswitch• Packetsthatdon’tconformmaybetreatedinoneofthefollowingways:– Givethepacketlowerprioritycomparedtopacketsinotheroutputqueues

– Labelthepacketasnonconformingbysettingtheappropriatebitsinaheader

– DiscardthepacketETSF05/ETSF10- InternetProtocols 50

TrafficManagement


• Fairness– Provideequaltreatmentofvariousflows

• Qualityofservice– Differenttreatmentfordifferentflows

• Reservations– Trafficcontractbetweenuserandnetwork– Excesstrafficdiscardedorhandledonabest-effortbasis

Supportfromroutingprotocols?

Yes!• Optimalpath

– Singlemetric– Multiplemetrics?

• Multiple Equal Cost Paths– Loadsharing– Loadbalancing

• QoSrouting– Cross-layerapproaches– OSPFextensions (RFC2676)

Well,sortof!• Appliestoalltraffic

– Nodifferentiation betweenflowtypes

• Whataboutinter-domain?– Nocontrolovernetwork

resources• Moresophisticated

mechanismsneeded– MultiprotocolLabelSwitching

(MPLS)– Resource Reservation


Layer 3congestion avoidance

• Congestion=dataload>networkcapacity– Arrival rate>processingrate– Processingrate>departurerate

• Asimplemethod– Randomearlydiscard(RED)– UDP?


APP

TCP

IP

Traffic descriptors


Traffic profiles


Trafficshaping:Twoapproaches

Leaky bucket Tokenbucket

56

Inputflow

Outputflow

Inputflow

Outputflow

ETSF05/ETSF10- InternetProtocols

Traffic shaping:Leakybucket


See also Figure 20.7

Traffic shaping:Tokenbucket


See also Figure 20.6

TokenBucket• Widelyusedtrafficmanagementtool• Advantages:–Manytrafficsourcescanbedefinedeasilyandaccurately

– Providesaconcisedescriptionoftheloadtobeimposedbyaflow,enablingtheservicetodetermineeasilytheresourcerequirement

– Providestheinputparameterstoapolicingfunction


Bonus:QoE,Quality of Experience

• Theuser’s subjective perceptionof thepresentationof thecontent

• Mean OpinionScore(MOS)• Researchforto find objective measures– Fullreference– Noreference– Hybrid

• What QoS give what QoE??


https://goo.gl/forms/IRcUYcJCZTu8uo152


Performance & QoS CongestionControl · Congestion avoidance Packet marking Traffic shaping Traffic...

Documents

Transcript of Performance & QoS CongestionControl · Congestion avoidance Packet marking Traffic shaping Traffic...