Lecture 8:

TCP and Congestion Control

Slides adapted from:Congestion slides for Computer Networks: A Systems Approach (Peterson and Davis)

Chapter 3 slides for Computer Networking: A Top Down Approach Featuring the Internet (Kurose and Ross)

ITCS 6166/8166 091Spring 2007

Jamie PaytonDepartment of Computer Science

University of North Carolina at Charlotte

February 5, 2007

Announcements

• Textbook is on reserve in library

• Homework 2 will be assigned on Wednesday– Due: Feb. 14

Transmission Control Protocol

• Implementation of sliding window protocol

TCP Segment (Packet) Structure

TCP uses sliding windows at sender and receiver

(buffers)

TCP Details

• Views data as ordered streams of bytes (not packets)– Sequence numbers are over bytes

• Seq # for segment = # of first byte

• Acknowledgements are cumulative– Acks for next expected byte #

• What happens on out-of-order segments?– Option 1: discard– Option 2: wait and see if other segments show up

• This approach taken in practice

TCP Sequence NumbersSeq. #’s:

– byte stream “number” of first byte in segment’s data

ACKs:

– seq # of next byte expected from other side

– cumulative ACK

Q: how receiver handles out-of-order segments

– A: TCP spec doesn’t say, - up to implementor

Host A Host B

Seq=42, ACK=79, data = ‘C’

Seq=79, ACK=43, data = ‘C’

Seq=43, ACK=80

Usertypes

‘C’

host ACKsreceipt

of echoed‘C’

host ACKsreceipt of

‘C’, echoesback ‘C’

timesimple telnet scenario

TCP: Sender• Assign sequence number to each segment (SeqNum)• Maintain three state variables:

– send window size (SWS) – upper bound on the number of outstanding (unACKed) frames that the sender can send

– last acknowledgment received (LAR)– last frame sent (LPS)

• Maintain invariant: LPS - LAR <= SWS

• Advance LAR when ACK arrives • Buffer up to SWS frames (in case retransmit required)• Timeout associated with each frame

SWS

LAR LFS

… …

TCP: Receiver• Maintain three state variables

– receive window size (RWS) – upper bound on the number of out-of-order frames the receiver can accept

– largest acceptable packet (LAP)– last packet received (LPR)

• Maintain invariant: LAP - LPR <= RWS

• Packet SeqNum arrives:– if LPR < SeqNum < = LAP accept– if SeqNum < = LPR or SeqNum > LAP discard

• Send cumulative ACKs

RWS

LPR LAP

… …

Buffer out of order packets!

TCP ACK generation [RFC 1122, RFC 2581]

Event at Receiver

Arrival of in-order segment withexpected seq #. All data up toexpected seq # already ACKed

Arrival of in-order segment withexpected seq #. One other segment has ACK pending

Arrival of out-of-order segmenthigher-than-expect seq. # .Gap detected

Arrival of segment that partially or completely fills gap

TCP Receiver action

Delayed ACK. Wait up to 500msfor next segment. If no next segment,send ACK

Immediately send single cumulative ACK, ACKing both in-order segments

Immediately send duplicate ACK, indicating seq. # of next expected byte

Immediate send ACK, provided thatsegment starts at lower end of gap

TCP Retransmission Timeout

• RTT estimate important for efficient operation– Too long: long delay before retransmission– Too short: unnecessary retransmission

• TCP RTT estimation– SampleRTT: measured time from segment

transmission until ACK receipt– SampleRTT will vary, want estimated RTT “smoother”

• average several recent measurements, not just current SampleRTT

TCP Round Trip Time

EstimatedRTT = (1- )*EstimatedRTT + *SampleRTT

• Exponential weighted moving average• influence of past sample decreases exponentially fast• typical value: = 0.125

Example RTT estimation:RTT: gaia.cs.umass.edu to fantasia.eurecom.fr

100

150

200

250

300

350

1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106

time (seconnds)

RTT

(mill

isec

onds

)

SampleRTT Estimated RTT

Setting the TCP Timeout• EstimtedRTT plus “safety margin”

– large variation in EstimatedRTT -> larger safety margin

• First estimate of how much SampleRTT deviates from EstimatedRTT:

TimeoutInterval = EstimatedRTT + 4*DevRTT

DevRTT = (1-)*DevRTT + *|SampleRTT-EstimatedRTT|

(typically, = 0.25)

Then set timeout interval:

Flow Control

• What happens if the receiving process is slow and the sending process is fast?

• TCP provides flow control– sender won’t overflow receiver’s buffer by

transmitting too much, too fast

How Flow Control Works

• Receiver advertises spare room to sender:RcvWindow = RcvBuffer-[LastByteRcvd-LastByteRead]

– includes RcvWindow in segments

• Sender keeps track of how much spare room receiver has in its variable RcvWindow

• Sender limits unACKed data to RcvWindow– guarantees receive buffer doesn’t overflow

Flow Control• Send buffer size: MaxSendBuffer• Receive buffer size: MaxRcvBuffer• Receiving side

– LastByteRcvd - LastByteRead < = MaxRcvBuffer– AdvertisedWindow = MaxRcvBuffer - (NextByteExpected - NextByteRead)

• Sending side– LastByteSent - LastByteAcked < = AdvertisedWindow– EffectiveWindow = AdvertisedWindow - (LastByteSent - LastByteAcked)

– LastByteWritten - LastByteAcked < = MaxSendBuffer– block sender if (LastByteWritten - LastByteAcked) + y > MaxSenderBuffer

• Always send ACK in response to arriving data segment• Persist when AdvertisedWindow = 0

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)

• A top-10 problem in computer network research!

TCP Congestion Control

• Idea– assumes best-effort network (FIFO or FQ routers)– each source determines network capacity for itself– uses implicit feedback– ACKs pace transmission (self-clocking)

• Challenges– determining the available capacity in the first place– adjusting to changes in the available capacity

TCP Congestion Control Fundamentals

• Additive Increase/Multiplicative Decrease

• Slow start

• Fast Retransmit and Fast Recovery

Additive Increase/Multiplicative Decrease

• Objective: adjust to changes in the available capacity• New state variable per connection: CongestionWindow

– Counterpart to flow control’s advertised window– limits how much data source has in transit

MaxWin = MIN(CongestionWindow, AdvertisedWindow)

EffWin = MaxWin - (LastByteSent - LastByteAcked)

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

• Now EffectiveWindow includes both flow control and congestion control

AIMD (cont)• 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

Source Destination

…

AIMD (cont)

• Algorithm– increment CongestionWindow

by one packet every time an entirewindow’s worth of data is successful (linear increase)

– divide CongestionWindow by two whenever a timeout occurs (multiplicative decrease)

– think about the concepts in termsof packets, even though its implemented in bytes

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

CongestionWindow += Increment

AIMD (cont)

• Trace: sawtooth behavior– proven to provide more stability than additive

increase/additive decrease– the consequences of having too large of a

window are much worse than having too small

60

20

1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0

KB

Time (seconds)

70

304050

10

10.0

Why AIMD?

• Fairness– If you run at any of the other combinations (AIAD, MIAD, MIMD),

common sequences of events can result in unfair distribution among competing flows

• R. Jain and K.K Ramakrishnan., Congestion Avoidance in Computer Networks with a Connectionless Network Layer: Concepts, Goals, and Methodology, in Proceedings of the Computer Networking Symposium, pp. 134—143, April 1988.

• Stability– If you’re slow starting, you KNOW that ½ of the window is OK,

so go back to there.– If you’re steady-state sending, then the reason your packet was

dropped is likely because a new flow started up• V. Jacobson., Congestion Avoidance and Control, in Proceedings of

the SIGCOMM Symposium, pp. 314—329, August 1988.

Slow Start

• Objective: determine the available capacity at first– increase congestion window rapidly

from a cold start

• Idea:– begin with CongestionWindow = 1

packet– double CongestionWindow each

RTT (increment by 1 packet for each ACK)

Source Destination

Slow Start (cont.)

• Why “slow”?– Starts slow in comparison to immediately

filling the advertised window– Prevents routers from having to handle bursts

of initial traffic.

Slow Start (cont)• when first starting connection

– the source has no idea what resources are available– Slow start continues to double CongWin until a loss occurs, then enters

additive increase/multiplicative decrease

• when connection goes dead waiting for timeout– receiver reopens entire window– use slow start to ramp up to previous CongWin/2

• Trace

• Problem: lose up to half a CongestionWindow’s worth of data– if you hit on the border of the network’s capacity (e.g., send n bytes worth of

data successfully, so double window to 2n, but network can only support n)

60

20

1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0

Time (seconds)

70

304050

10

Fast Retransmit and Fast Recovery

• Problem: coarse-grain TCP timeouts lead to idle periods

• Fast retransmit: use duplicate ACKs to trigger retransmission faster than normal– does not replace regular timeout

but enhances it– duplicate ACK suggests

problem at receiver• three duplicate ACKs triggers a

retransmit for the missing packet

Packet 1

Packet 2

Packet 3

Packet 4

Packet 5

Packet 6

Retransmitpacket 3

ACK 1

ACK 2

ACK 2

ACK 2

ACK 6

ACK 2

Sender Receiver

Fast Retransmit Results

• no more long, flat periods where no packets are sent, waiting for a timeout– eliminates about half of the timeouts in practice

• Fast recovery– skip the slow start phase– go directly to half the last successful CongestionWindow

(ssthresh)

60

20

1.0 2.0 3.0 4.0 5.0 6.0 7.0

Time (seconds)

70

304050

10

Congestion Avoidance• TCP’s strategy

– control congestion once it happens– repeatedly increase load in an effort to find the point at

which congestion occurs, and then back off• i.e., cause congestion in order to control it

• Alternative strategy– predict when congestion is about to happen– reduce rate before packets start being discarded– call this congestion avoidance, instead of congestion control

TCP Vegas

• End host management of congestion avoidance

• Watch for (implicit) signs that congestion is building– e.g. steady increase in RTT, flattening of the

perceived throughput

TCP Vegas• While we’re

increasing the congestion window, the perceived throughput stays about the same– i.e., the extra

stuff is being queued

– TCP Vegas tries to reduce the amount of this extra stuff

60

20

0.5 1.0 1.5 4.0 4.5 6.5 8.0

KB

Time (seconds)

Time (seconds)

70

304050

10

2.0 2.5 3.0 3.5 5.0 5.5 6.0 7.0 7.5 8.5

900

300100

0.5 1.0 1.5 4.0 4.5 6.5 8.0

Sen

ding

KB

ps1100

500700

2.0 2.5 3.0 3.5 5.0 5.5 6.0 7.0 7.5 8.5

Time (seconds)0.5 1.0 1.5 4.0 4.5 6.5 8.0Q

ueue

siz

e in

rou

ter

5

10

2.0 2.5 3.0 3.5 5.0 5.5 6.0 7.0 7.5 8.5

Algorithm • Let BaseRTT be the minimum of all measured RTTs (commonly the RTT

of the first packet)• If not overflowing the connection, then

ExpectedRate = CongestionWindow/BaseRTT• Source calculates sending rate (ActualRate) once per RTT

– Record the send time for a particular packet, record how many bytes are transmitted before the ack comes, compute the sample RTT for that packet, and divide the number of bytes transmitted in the middle by the RTT

• Source compares ActualRate with ExpectedRate

Diff = ExpectedRate - ActualRateif Diff <

increase CongestionWindow linearly (not using bandwidth effectively)

else if Diff > decrease CongestionWindow linearly(moving towards congestion)

elseleave CongestionWindow unchanged

Algorithm (cont)

• Parameters(in practice) = 1 packet = 3 packets

70605040302010

KB

Time (seconds)

0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0

0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0

CA

M K

Bps

240200160120

8040

Time (seconds)