SMU CSE 5344/7344 1

Congestion Control

Class 18Resource Allocation

Considerations

SMU CSE 5344/7344 2

Congestion Control

• Issues and Principles• Causes of Congestion• Approaches to Congestion

Control

SMU CSE 5344/7344 3

Terminology

• Resource Allocation– Process by which NEs try to meet

competing demands from the same network resources:

• Bandwidth + buffer space

• Congestion Control– Efforts made by NEs to prevent or

respond to overload conditions in the NW

– Could demand certain NEs to stopTXing– Prefer a notion of fairness

SMU CSE 5344/7344 4

Issues – Congestion on Switched NWs

• Two sides of the same coin– pre-allocate resources so at to avoid congestion– control congestion if (and when) is occurs

• Two points of implementation– hosts at the edges of the network (transport

protocol)– routers inside the network (queuing discipline)

• Underlying service model– best-effort (assume for now)– multiple qualities of service (later)– soft-state versus hard-state flow information

Destination1.5-Mbps T1 link

Router

Source2

Source1

100-Mbps FDDI

10-Mbps Ethernet

Lots of capacity on immediate links

Reduced capacity on NW links

“Default Router” – the bottleneck

SMU CSE 5344/7344 5

Framework

• Taxonomy– router-centric versus host-centric– reservation-based versus feedback-based

• Implicit (behavioral) vs. explicit feedback– window-based versus rate-based

Router

Source2

Source1

Source3

Router

Router

Destination2

Destination1

Data “flows” – visible to routers inside the network

Channels – process-to-process

SMU CSE 5344/7344 6

Principles of Congestion Control

Congestion:

• informally: “too many sources sending too much data too fast for network to handle”

– different from flow control

• manifestations:

– lost packets (buffer overflow at routers)

– long delays (queueing in router buffers)

SMU CSE 5344/7344 7

Congestion: A Close-up View

• knee – point after which – throughput increases

very slowly– delay increases fast

• cliff – point after which– throughput starts to

decrease very fast to zero (congestion collapse)

– delay approaches infinity

Load

De

lay

cliff

congestioncollapse

Load

Th

rou

ghp

ut knee packet

loss

SMU CSE 5344/7344 8

Congestion Control vs. Congestion Avoidance

• Congestion control goal– stay left of cliff

• Congestion avoidance goal– stay left of knee

• Right of cliff: – Congestion collapse

Load

Th

rou

ghp

ut knee cliff

congestioncollapse

SMU CSE 5344/7344 9

Congestion Collapse: How Bad is It?

• Definition: Increase in network load that results in a decrease of useful work done

• Many possible causes– Spurious retransmissions of packets still in flight– Undelivered packets

• Packets consume resources and are dropped elsewhere in network

– Fragments• Mismatch of transmission and retransmission units

– Control traffic• Large percentage of traffic is for control

– Stale or unwanted packets• Packets that are delayed on long queues

SMU CSE 5344/7344 10

Evaluation Criteria for Resource Allocation Scheme Effectiveness

• Fairness• Power (ratio of throughput to

delay)

Optimalload Load

Th

rou

ghp

ut/d

elay

Th

rou

gh

pu

t

Dela

y

SMU CSE 5344/7344 11

Causes of Congestion

SMU CSE 5344/7344 12

Causes/costs of congestion: scenario 1

• two senders, two receivers

• one router, infinite buffers

• no retransmission

• large delays when congested

• maximum achievable throughput

unlimited shared output link buffers

Host A

in : original

data

Host B

out

Per-connection throughput

C= Router Capacity

Aggregate throughput = C is bad for delay

SMU CSE 5344/7344 13

Causes/costs of congestion: scenario 2 • one router, finite buffers • sender retransmission of lost packet• Rate of sending original data +

retransmissions offered load

finite shared output link buffers

Host A in : original data

Host B

out

'in : original data, plus retransmitted data

'in

SMU CSE 5344/7344 14

Causes/costs of congestion: scenario 2

• always: (goodput)

• “perfect” retransmission only when loss:

• retransmission of delayed (not lost) packet makes

larger (than perfect case) for same

in

out

=

in

out

>

in

out

“costs” of congestion: • more work (retrans) for given “goodput”• unneeded retransmissions: link carries multiple copies of pkt

No Loss case

Retransmissions due to loss

Retransmissions due to queue

delayC/4

SMU CSE 5344/7344 15

finite shared output link buffers

Host A in : original data

Host D

out

'in : original data, plus retransmitted data

R1

R2

R3

R4

Host B

Host C

Causes/costs of congestion scenario 3

• four senders• multihop paths• timeout/retransmit

Q: what happens as and increase ?

in

Max R2 Capacity = C

If 'in is very large, arrival rate at R2 can overwhelm R2

As offered load infinity, B-D traffic supersedes A-C traffic to zero

in

SMU CSE 5344/7344 16

Causes/costs of congestion: scenario 3

Another “cost” of congestion:

• when packet dropped, any “upstream transmission capacity” used for that packet is wasted

NW is better off if R1 discarded packets NW is better off if R2 preferentially transmitted packets

that already crossed the network

Host A

Host B

o

u

t

Offered Load

Th

rou

gh

pu

t

Multi-hop flows vs. single-hop flows fair?

SMU CSE 5344/7344 17

Resource Allocation Schemes

SMU CSE 5344/7344 18

Queuing Discipline

• First-In-First-Out (FIFO)– does not discriminate between traffic sources (flows)

– Tail-drop – is a drop policy (not a scheduling method)

– Pushes congestion control / resource allocation responsibility out to the hosts

SMU CSE 5344/7344 19

Queuing Discipline• FIFO with Priority Queuing (FIFO-PQ)

– Multiple FIFO queues at each router

– Differential (TOS-based) drops

• Lets VIP jump to head of the line

• Problem?– Can “starve-out” lower-priority packets

– Needs a means to force those packets out as well

SMU CSE 5344/7344 20

Queuing Discipline• Fair Queuing (FQ)

– explicitly segregates traffic based on flows– ensures no flow captures more than its share of

capacity• punishes non-conforming flows

– variation: weighted fair queuing (WFQ)– still needs an end-to-end congestion control

mechanism

• Problem?Flow 1

Flow 2

Flow 3

Flow 4

Round-robinservice

Different Packet lengths

Algorithm “views transmission” as bit-by-bit

SMU CSE 5344/7344 21

FQ Algorithm – single flow

• Suppose clock ticks each time a bit is transmitted

• Let Pi denote the length of packet i

• Let Si denote the time when start to transmit packet i

• Let Fi denote the time when finish transmitting packet i

• Fi = Si + Pi

• When does router start transmitting packet i?– if before router finished tx packet i - 1 from this flow,

then tx immediately after last bit of i - 1 (Fi-1)

– if no current packets for this flow, then start transmitting when arrives (call this Ai)

• Thus: Fi = MAX (Fi - 1, Ai) + Pi

SMU CSE 5344/7344 22

FQ Algorithm – multiple flows• For multiple flows

– calculate Fi for each packet that arrives on each flow

– treat all Fi’s as timestamps

– next packet to transmit is one with lowest timestamp• The one that should finish before the others based on

time

• Not perfect: can’t preempt current packet– So not a perfect case of bit-by-bit round robin

• ExampleFlow 1 Flow 2

(a) (b)

Output Output

F = 8 F = 10F = 5

F = 10

F = 2

Flow 1(arriving)

Flow 2(transmitting)

Select F5,F8 before F10 F2 is smaller and finished before F10, if true bit-by-bit round robin

SMU CSE 5344/7344 23

FQ Algorithm – multiple flows• Link is never left idle – work conserving

– Active NE can use full link capacity

• One flow cannot use more than 1/n capactiy if have n flows– Provides a guaranteed minimum shared bandwidth is

provided each flow

• Weighted Fair Queuing– Specifies (weighs) how much bandwidth any flow

receives– FQ, weight = 1 for each flow– Mechanism can provide TOS

General system concept separate Mechanism from Policy

SMU CSE 5344/7344 24

Approaches to Congestion Control

SMU CSE 5344/7344 25

Approaches towards congestion control

End-end congestion control:

• no explicit feedback from network– To Transport Layer

• congestion inferred from end-system observed loss, delay

• approach taken by TCP– Since IP does not

provide any feedback

Network-assisted congestion control:

• routers (network-layer components) provide feedback to end systems

• Feedback may be:– single bit indicating

congestion (SNA, DECnet, ATM-ABR)

a router tells the sender the explicit rate to transmit on a link

Two broad approaches towards congestion control:

SMU CSE 5344/7344 26

Network-assisted Congestion Control:ATM ABR

ABR: available bit rate:

• “elastic service” if sender’s path “under-loaded”: – sender should use

available bandwidth• if sender’s path

congested: – sender throttled to

minimum guaranteed rate

RM (resource management) cells:

• sent by sender, interspersed with data cells

• bits in RM cell set by switches (“network-assisted”)

• RM cells returned to sender by receiver, with bits intact

SMU CSE 5344/7344 27

ATM ABR Congestion Control

ABR Congestion Control is a rate-based approach:

• Sender computes a maximum rate at which it can send• Sender regulates transmissions accordingly

Three mechanisms to signal congestion-related info:

1. bits in RM cell set by switches (“network-assisted”) – RM cells are interspersed with data cells (1:32

default)– NI bit: no increase in rate (mild congestion) – bit = 1– CI bit: congestion indication – bit = 1– Switches set these bits as RM cells arrive

SMU CSE 5344/7344 28

ATM ABR Congestion ControlThree mechanisms to signal congestion-related

info:

• ER Setting – Explicit Rate– Each RM cell also contains a two-byte ER field– Congested switch lowers the value contained in the ER

field of a passing RM cell– This sets the minimum rate is across all src<>dst

switches

• EFCI bit – Explicit Forward Congestion Indication– Each data cell contains an EFCI bit– Switch sets EFCI bit = 1 to inform destination host– Destination sets CI bit = 1 in RM cell to source host

ATM ABR source adjusts sending rate as a function of CI, NI, ER values

SMU CSE 5344/7344 29

End-to-End Congestion Control

SMU CSE 5344/7344 30

TCP Congestion Control• Idea

– assumes best-effort network (FIFO or FQ routers) each source determines network capacity for itself

– uses implicit feedback– the ACKs pace transmission (self-clocking)

• Challenge– determining the available capacity in the

first place– adjusting to changes in the available

capacity

SMU CSE 5344/7344 31

Additive Increase/Multiplicative Decrease (AIMD)

Objective: adjust to changes in the available capacity

• New state variable per connection: CongestionWindow– limits how much data source has in transit– NW congestion counterpart of flow-control

AdvertisedWindow

MaxWin = MIN(CongestionWindow,AdvertisedWindow) EffWin = MaxWin - (LastByteSent - LastByteAcked)

• Idea:– increase CongestionWindow when congestion goes down– decrease CongestionWindow when congestion goes up

SMU CSE 5344/7344 32

TCP Congestion ControlHow does sender perceive congestion?• loss event = timeout or 3 duplicate ACKs• TCP sender reduces rate (CongWin) after loss

event

three mechanisms to adjust:– AIMD– slow start– conservative after timeout events

SMU CSE 5344/7344 33

AIMD (cont’d)

Question: how does the source determine whether or not the network is congested?

• Answer: a timeout occurs– timeout signals that a packet was lost– packets are seldom lost due to transmission error– lost packet implies congestion

• Behaviorally based

SMU CSE 5344/7344 34

AIMD (cont)

• In practice: increment a little for each ACKIncrement = MSS * (MSS/CongestionWindow)

CongestionWindow += Increment

Source Destination

…

• Algorithm– increment CongestionWindow by one

packet per RTT (linear increase)– divide CongestionWindow by two

whenever a timeout occurs (multiplicative decrease)

SMU CSE 5344/7344 35

TCP AIMDmultiplicative decrease:

cut CongWin in half after loss event

additive increase: increase CongWin by 1 MSS every RTT in the absence of loss events: probing

8 Kbytes

16 Kbytes

24 Kbytes

time

congestionwindow

Long-lived TCP connection

AIMD is necessary for a stability

Decrease aggressively / increase conservatively consequences are less for a smaller window than a larger window

SMU CSE 5344/7344 36

TCP Slow Start

• When connection begins, CongWin = 1 MSS– Example: MSS = 500

bytes & RTT = 200 msec– initial rate = 20 kbps

• available bandwidth may be >> MSS/RTT– desirable to quickly ramp

up to respectable rate

• When connection begins, increase rate exponentially fast until first loss event

SMU CSE 5344/7344 37

TCP Slow Start (more)• When connection

begins, increase rate exponentially until first loss event:– double CongWin every

RTT– done by incrementing CongWin for every ACK received

• Summary: initial rate is slow but ramps up exponentially fast

Host A

one segment

RTT

Host B

time

two segments

four segments

SMU CSE 5344/7344 38

Slow Start (cont’d)

• Exponential growth, but slower than all at once

• Used…– when first starting connection– when connection goes dead waiting for

timeout

SMU CSE 5344/7344 39

Summary: TCP Congestion Control

• When CongWin is below Threshold, sender in slow-start phase, window grows exponentially

• When CongWin is above Threshold, sender is in congestion-avoidance phase, window grows linearly

• When a triple duplicate ACK occurs, Threshold set to CongWin/2 and CongWin set to Threshold

• When timeout occurs, Threshold set to CongWin/2 and CongWin is set to 1 MSS.

SMU CSE 5344/7344 40

TCP sender congestion control

Event State TCP Sender Action Commentary

ACK receipt for previously unACKed data

Slow Start (SS)

CongWin = CongWin + MSS, If (CongWin > Threshold) set state to “Congestion Avoidance”

Resulting in a doubling of CongWin every RTT

ACK receipt for previously unacked data

CongestionAvoidance (CA)

CongWin = CongWin+MSS * (MSS/CongWin)

Additive increase, resulting in increase of CongWin by 1 MSS every RTT

Loss event detected by triple duplicate ACK

SS or CA Threshold = CongWin/2, CongWin = Threshold,Set state to “Congestion Avoidance”

Fast recovery, implementing multiplicative decrease. CongWin will not drop below 1 MSS.

Timeout SS or CA Threshold = CongWin/2, CongWin = 1 MSS,Set state to “Slow Start”

Enter slow start

Duplicate ACK SS or CA Increment duplicate ACK count for segment being ACKed

CongWin and Threshold not changed

SMU CSE 5344/7344 41

End of Class 18