13 congestion in data networks real
TRANSCRIPT
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 1/47
Data and Computer Data and Computer
CommunicationsCommunications
Ninth EditionNinth Edition
by William Stallingsby William Stallings
Chapter 13Chapter 13 ± ± Congestion in DataCongestion in DataNetworksNetworks
Data and Computer Communications, NinthData and Computer Communications, NinthEdition by William Stallings, (c) PearsonEdition by William Stallings, (c) Pearson
EducationEducation -- Prentice Hall, 2011Prentice Hall, 2011
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 2/47
Congestion in Data NetworksCongestion in Data Networks
We were doing very well, up to the kind of sum when a bath is filling at theWe were doing very well, up to the kind of sum when a bath is filling at therate of so many gallons and two holes are letting the water out, and pleaserate of so many gallons and two holes are letting the water out, and please
to say how long it will take to fill the bath, when my mother put down theto say how long it will take to fill the bath, when my mother put down the
socks she was darning and clicked her tongue in impatience.socks she was darning and clicked her tongue in impatience.
"Filling up an old bath with holes in it, indeed. Who would be such a fool?" "Filling up an old bath with holes in it, indeed. Who would be such a fool?"
"A sum it is, girl," my father said. "A sum. A problem for the mind." "A sum it is, girl," my father said. "A sum. A problem for the mind." "Filling the boy with old nonsense," Mama said."Filling the boy with old nonsense," Mama said.
"Not nonsense, Beth," my father said. "A sum, it is. The water pours in and "Not nonsense, Beth," my father said. "A sum, it is. The water pours in and
takes so long. It pours out and takes so long. How long to fill? That is all." takes so long. It pours out and takes so long. How long to fill? That is all."
"But who would pour water into an old bath with holes?" my mother said."But who would pour water into an old bath with holes?" my mother said.
"Who would think to do it, but a lunatic?" "Who would think to do it, but a lunatic?"
²²How Green Was My Valley,How Green Was My Valley,
Richard LlewellynRichard Llewellyn
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 3/47
What Is Congestion?What Is Congestion?
congestion occurs when thecongestion occurs when the number number of packetsof packets
being transmitted through the networkbeing transmitted through the network
approaches the packet handling capacity of theapproaches the packet handling capacity of the
networknetwork
oobjective of congestionbjective of congestion controlicontroliss to keepto keep thethe
number number of packets below a level at whichof packets below a level at which
performance falls off dramaticallyperformance falls off dramatically a data network is a network of queuesa data network is a network of queues
generally 80% utilization is criticalgenerally 80% utilization is critical
finite queues mean data may be lostfinite queues mean data may be lost
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 4/47
Queues at a NodeQueues at a Node
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 5/47
Interaction of QueuesInteraction of Queues
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 6/47
IdealIdeal
NetworkNetworkUtilizationUtilization
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 7/47
Effects of Effects of
CongestionCongestion--
No ControlNo Control
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 8/47
Mechanisms for Mechanisms for
Congestion ControlCongestion Control
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 9/47
BackpressureBackpressure
if node becomes congested it can slow down or if node becomes congested it can slow down or halt flow of packets from other nodeshalt flow of packets from other nodes same effect as backpressure in a blocked fluid pipesame effect as backpressure in a blocked fluid pipe
other nodes have to apply control on incoming packetother nodes have to apply control on incoming packet
ratesrates propagates back to sourcepropagates back to source
can restrict to high traffic logical connectionscan restrict to high traffic logical connections
used in connection oriented networks that allowused in connection oriented networks that allow
hophop--byby--hop congestion control (eg. X.25)hop congestion control (eg. X.25) not used in ATM or frame relaynot used in ATM or frame relay
only recently developed for IPonly recently developed for IP
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 10/47
Choke PacketChoke Packet
a controla control packetpacket
generated at congested nodegenerated at congested node
sent back to source nodesent back to source node
Internet Control Message Protocol (ICMP) SourceInternet Control Message Protocol (ICMP) Source
Quench packetQuench packet
�� from router or destination end systemfrom router or destination end system
�� source cuts back until it no longer receives quench messagessource cuts back until it no longer receives quench messages
�� message is issued for every discarded packetmessage is issued for every discarded packet�� message may also be issued for anticipated congestionmessage may also be issued for anticipated congestion
crude control techniquecrude control technique
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 11/47
Implicit Congestion SignalingImplicit Congestion Signaling
with network congestion:with network congestion:
ttransmission delay increasesransmission delay increases
ppackets may be discardedackets may be discarded
source can detect congestion and reduce flowsource can detect congestion and reduce flow
r responsibility of end systemsesponsibility of end systems
effectiveeffective on connectionless (datagram)on connectionless (datagram) networksnetworks
aalso used in connectionlso used in connection--oriented networksoriented networks LAPF control is capable of detecting lost framesLAPF control is capable of detecting lost frames
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 12/47
Expli it ti i liExpli it ti i li
BackwardsBackwards
congestion avoidancecongestion avoidance
notification in oppositenotification in opposite
direction to packetdirection to packetrequiredrequired
ForwardsForwards
congestion avoidancecongestion avoidance
notification in samenotification in samedirection as packetdirection as packet
requiredrequired
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 13/47
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 14/47
Tr ffi M tTr ffi M t
fairness
� provide equal treatment of various flows
quality of service� different treatment for different connections
reservations� traffic contract between user and network
� e cess traffic discarded or handled on a best-effort basis
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 15/47
Congestion Control in PacketCongestion Control in Packet
Switched NetworksSwitched Networks send control packet to some or all sourcesend control packet to some or all source
nodesnodes
requires additional traffic during congestionrequires additional traffic during congestion
rely on routing informationrely on routing information
may react too quicklymay react too quickly
end to end probe packetsend to end probe packets
adds to overheadadds to overhead
add congestionadd congestion informationinformation to packets into packets intransittransit
either backwards or forwardseither backwards or forwards
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 16/47
A M Traffic ManagementATM Traffic Management
high speed, small cell size, limited overhead bitshigh speed, small cell size, limited overhead bits
stillstill evolvingevolving
reasonsreasonse istinge istingtools aretools areinadequateinadequatefor ATMfor ATM
majority of traffic not amenable to flow control
feedback is slow due to reduced transmission time comparedwith propagation delay
wide range of application demands
different traffic patterns
different network services
high speed switching and transmission increases volatility
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 17/47
Latency/Speed EffectsLatency/Speed Effects� consider ATM at
150Mbps
� takes ~2.8x10-6
seconds to insert asingle cell
� time to traversenetwork dependson propagationdelay andswitching delay
� assume propagationat two-thirds speedof light
� if source anddestination on
opposite sides of USA, propagationtime ~ 48x10-3seconds
� given implicitcongestion control, by the timedropped cellnotification has
reached source, 7.2x106 bits havebeen transmitted
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 18/47
Cell Delay VariationCell Delay Variation
for ATM voice/video, data is a stream of cellsfor ATM voice/video, data is a stream of cells
delay across network must be shortdelay across network must be short
rate of delivery must be constantrate of delivery must be constant there will always be some variation in transitthere will always be some variation in transit
delay cell delivery to application so that constantdelay cell delivery to application so that constant
bit rate can be maintained to applicationbit rate can be maintained to application
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 19/47
Timing of CBR CellsTiming of CBR Cells
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 20/47
Network Contribution toNetwork Contribution to
Cell Delay VariationCell Delay Variation in packet switched networks is due toin packet switched networks is due to
queuingqueuing delaysdelays and routing decisionand routing decision timetime
in ATM networksin ATM networks less than frame relayless than frame relay
ATM protocol designed to minimize ATM protocol designed to minimizeprocessingprocessing overheadoverhead at switchesat switches
ATM switches have very high throughput ATM switches have very high throughput
only noticeable delay is from congestiononly noticeable delay is from congestion
must not accept load that causes congestionmust not accept load that causes congestion
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 21/47
Cell Delay VariationCell Delay Variation
At The UNIAt The UNI
cannot predict these delayscannot predict these delays
3 layers of ATM processing causes delay
� interleaving cells from different connections
� operation and maintenance cell interleaving
� if using synchronous digital hierarchy frames,these are inserted at physical layer
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 22/47
Origins of Cell Delay VariationOrigins of Cell Delay Variation
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 23/47
Traffic and CongestionTraffic and Congestion
Control FrameworkControl Framework
ATM layer traffic and congestion control ATM layer traffic and congestion controlshould support QoS classes for allshould support QoS classes for allforeseeable network servicesforeseeable network services
should not rely on AAL protocols that areshould not rely on AAL protocols that arenetwork specific, nor on higher levelnetwork specific, nor on higher level
application specific protocolsapplication specific protocols should minimize network and end to endshould minimize network and end to end
system comple itysystem comple ity
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 24/47
Traffic Control andTraffic Control and
Congestion Control FunctionsCongestion Control Functions
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 25/47
Traffic Management andTraffic Management and
Congestion ControlCongestion Control
TechniquesTechniques
resource management using virtual pathsresource management using virtual paths
connection admission controlconnection admission control usage parameter controlusage parameter control
selective cell discardselective cell discard
traffic shapingtraffic shaping
ATM traffic ATM traffic
techniques:techniques:
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 26/47
Resource Management UsingResource Management Using
Virtual PathsVirtual Paths QoS parametersQoS parameters
concerned with are:concerned with are: cell loss ratiocell loss ratio cell transfer delaycell transfer delay
cell delay variationcell delay variation
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 27/47
Configuration of Configuration of
VCCs and VPCsVCCs and VPCs
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 28/47
Allocating VCCs within VPCAllocating VCCs within VPC
all VCCs within VPC should e perienceall VCCs within VPC should e periencesimilar network performancesimilar network performance
options for allocationoptions for allocation::
aggregatepeak
demand
� set VPC capacity to total of all peakVCC rates
� will meet peak demands, but oftenunderutilized
statisticalmultiple ing
� set VPC capacity to more thanaverage VCC rates
� will see greater variation but better utilization
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 29/47
Connection AdmissionConnection Admission
ControlControl first line of defensefirst line of defense
user specifies traffic characteristics for newuser specifies traffic characteristics for newconnection (VCC or VPC) by selecting a QoSconnection (VCC or VPC) by selecting a QoS
network accepts connection only if it can meetnetwork accepts connection only if it can meetthethe demand and maintain QoSdemand and maintain QoS
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 30/47
Traffic Parameters Used inTraffic Parameters Used in
Defining VCC/VPC Quality of Defining VCC/VPC Quality of
ServiceService
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 31/47
Procedures Used to Set Values of Procedures Used to Set Values of
Traffic Contract ParametersTraffic Contract Parameters
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 32/47
Traffic Rate ManagementTraffic Rate Management
must discard frames to cope with congestionmust discard frames to cope with congestion
arbitrarily, no regard for sourcearbitrarily, no regard for source
no reward for restraint so end systems transmit asno reward for restraint so end systems transmit as
fast as possiblefast as possible ccommittedommitted information rate (CIR)information rate (CIR)
�� data in e cess of this liable to discarddata in e cess of this liable to discard
�� not guaranteed in e treme congestion situationsnot guaranteed in e treme congestion situations
�� aggregate CIR should not e ceed physical data rateaggregate CIR should not e ceed physical data rate
ccommittedommitted burst sizeburst size
ee cesscess burst sizeburst size
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 33/47
Usage Parameter Usage Parameter ControlControl
(UPC)(UPC)
UPC function monitors a connection toUPC function monitors a connection to
ensure traffic obeys contractensure traffic obeys contract
purpose is to protect network resourcespurpose is to protect network resources
from overloadfrom overload onon one connectionone connection
done on VCC and VPCdone on VCC and VPC
peak cell rate and cell delay variationpeak cell rate and cell delay variation
sustainable cell rate and burstsustainable cell rate and burst tolerancetolerance
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 34/47
Cell R te Al ritCell R te Al rit
ttrafficraffic flow is compliant if the peak rate of flow is compliant if the peak rate of
cell transmission does not e ceed thecell transmission does not e ceed the
agreed peak cell rateagreed peak cell rate ccellell rate algorithm:rate algorithm:
known as theleaky bucket
algorithm
defines trafficcompliance
compliantcells are
passed on
noncompliantcells are
discarded
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 35/47
SelectiveSelective CellCell DiscardDiscard
comes in to play when network at a pointcomes in to play when network at a point
beyond UPC discards (CLP=1) cellsbeyond UPC discards (CLP=1) cells
object is to discard lower object is to discard lower--priority cellspriority cellswhen congested to protect higher when congested to protect higher--prioritypriority
cellscells
network has no way to discriminatenetwork has no way to discriminate
between cells that were labeled as lower between cells that were labeled as lower
priority by the source and that were taggedpriority by the source and that were tagged
by the UPC functionby the UPC function
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 36/47
Traffic ShapingTraffic Shaping
UPC referred to as traffic policingUPC referred to as traffic policing
cells that e ceed a certain performance levelcells that e ceed a certain performance level
are discarded or taggedare discarded or tagged
can be desirable to also shape trafficcan be desirable to also shape traffic
traffic shaping reduces cell clumpingtraffic shaping reduces cell clumping
token buckettoken bucket
controls flow of compliant cellscontrols flow of compliant cells
smoothes out bursts of cellssmoothes out bursts of cells
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 37/47
Token BucketToken Bucket for for
Traffic ShapingTraffic Shaping
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 38/47
GFR Traffic ManagementGFR Traffic Management
guaranteed frame rate (GFR) as simple as UBRguaranteed frame rate (GFR) as simple as UBRfrom end system viewpointfrom end system viewpoint
places modest requirements on ATM networkplaces modest requirements on ATM network
end system does no policing or shaping of trafficend system does no policing or shaping of traffic may transmit at line rate of ATM adaptor may transmit at line rate of ATM adaptor
no guarantee of deliveryno guarantee of delivery so higher layer (eg. TCP) must do congestion controlso higher layer (eg. TCP) must do congestion control
user can reserve capacity for each VCuser can reserve capacity for each VC ensures application can transmit at minimum rate withensures application can transmit at minimum rate with
no lossno loss
if no congestion, higher rates may be usedif no congestion, higher rates may be used
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 39/47
GFR Contract ParametersGFR Contract Parameters
Peak cell rate (PCR)Peak cell rate (PCR)
Minimum cell rate (MCR)Minimum cell rate (MCR)
Ma imum burst size (MBS)Ma imum burst size (MBS)Ma imum frame size (MFS)Ma imum frame size (MFS)
Cell delay variation tolerance (CDVT)Cell delay variation tolerance (CDVT)
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 40/47
Components of GFR SystemComponents of GFR System
SupportingR
ateG
uaranteesSupportingR
ateG
uarantees
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 41/47
Tagging and PolicingTagging and Policing
discriminates between frames thatdiscriminates between frames that
conform to contract and those that don¶tconform to contract and those that don¶t
set CLP=1 on all cells of each frame thatset CLP=1 on all cells of each frame thatdoes not conformdoes not conform
gives lower prioritygives lower priority
tagging can be done by network or sourcetagging can be done by network or source network may discard CLP=1 cellsnetwork may discard CLP=1 cells
ccellell discard considered a policing functiondiscard considered a policing function
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 42/47
Buffer ManagementBuffer Management
deals with treatment of buffered cellsdeals with treatment of buffered cells
congestion indicated by high buffer congestion indicated by high buffer occupancyoccupancy
will discard tagged cells in preference towill discard tagged cells in preference tountagged cellsuntagged cells
including ones already in buffer to make roomincluding ones already in buffer to make room
may do per VC buffering for fairnessmay do per VC buffering for fairness cell discard based on queuecell discard based on queue--specificspecific
thresholdsthresholds
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 43/47
SchedulingScheduling
preferential treatment to untagged cellspreferential treatment to untagged cells
separate queues for each VCseparate queues for each VC
make per make per--VC scheduling decisionsVC scheduling decisions enables control of outgoing rate of VCsenables control of outgoing rate of VCs
VCs get fair capacity allocationVCs get fair capacity allocation
still meet contractstill meet contract
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 44/47
GFC Conf or nce efinitionGFC Conf or nce efinition
UPC functionUPC function
monitors each activemonitors each active
VC to ensure trafficVC to ensure traffic
conforms to contractconforms to contractand tags or discardsand tags or discards
nonconforming cellsnonconforming cells
frame conforms if allframe conforms if all
cells conformcells conform
all cells in framehave same CLP
rate of cells iswithin contract
frame satisfiesMFS parameter
� check if either
last cell in frameor cell count <MFS
a cell conforms
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 45/47
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 46/47
QoS Eligibility TestQoS Eligibility Test
two stage filtering processtwo stage filtering process
a frame is tested for conformance to contracta frame is tested for conformance to contract may discard or tagmay discard or tag
set upper bound & penalize cells above upper boundset upper bound & penalize cells above upper bound do e pect attempt to deliver tagged cellsdo e pect attempt to deliver tagged cells
determine frames eligible for QoS guaranteesdetermine frames eligible for QoS guarantees under GFR contract for VCunder GFR contract for VC
set lower bound on trafficset lower bound on traffic frames in traffic flow below threshold are eligibleframes in traffic flow below threshold are eligible
8/6/2019 13 Congestion in Data Networks REAL
http://slidepdf.com/reader/full/13-congestion-in-data-networks-real 47/47
SummarySummary
congestion effectscongestion effects ideal and practical performanceideal and practical performance
congestioncongestion controlcontrol
bbackpressure, choke packet, implicit/e plicitackpressure, choke packet, implicit/e plicit traffictraffic managementmanagement
f fairness, QoS, reservationsairness, QoS, reservations
ATM traffic management ATM traffic management ATM ATM--GFR traffic managementGFR traffic management
tagging, policing, buffer, schedulingtagging, policing, buffer, scheduling
QoS eligibility testingQoS eligibility testing