advanced networks 20021 transport layer - ii michalis faloutsos many slides from kurose-ross...

Advanced Networks 2002

1

Transport Layer - II

Michalis FaloutsosMichalis Faloutsos

Many slides from Kurose-RossMany slides from Kurose-Ross

Srikanth KrishnamurthySrikanth Krishnamurthy

S. KalyanaramanS. Kalyanaraman


2

In this set of slides

TCP variantsTCP variants

Router assisted congestion controlRouter assisted congestion control


3

Current TCP Versions

TCP specs can be implemented in different TCP specs can be implemented in different waysways

TCP versions:TCP versions:• Tahoe (basic as described)Tahoe (basic as described)• RenoReno• Las VegasLas Vegas


4

TCP Reno

Most popular TCP implementationMost popular TCP implementationFast retransmit on 3 duplicate ACKsFast retransmit on 3 duplicate ACKsFast recovery: cancel slow start after fast Fast recovery: cancel slow start after fast retransmissionretransmission• Half the congestion window threshold, but start with Half the congestion window threshold, but start with

congestion window equal to thresholdcongestion window equal to threshold• Go to congestion avoidance phaseGo to congestion avoidance phase

Optimistic Rationale: Optimistic Rationale: I hope there was only one packet lostI hope there was only one packet lost Since I sent it, I hope it arrives this timeSince I sent it, I hope it arrives this time


5

Fast RetransmitCoarse grained TCP time-outs sometimes lead to long periods Coarse grained TCP time-outs sometimes lead to long periods wherein a connection goes dead waiting for a timer to expire.wherein a connection goes dead waiting for a timer to expire.Fast Retransmit -- a heuristic that sometimes “triggers” the Fast Retransmit -- a heuristic that sometimes “triggers” the retransmission of a packet faster than permissible by the regular retransmission of a packet faster than permissible by the regular time-out.time-out.Every time a data packet arrives at a receiver, the receiver ACKs Every time a data packet arrives at a receiver, the receiver ACKs even though the particular sequence number has been ACKed.even though the particular sequence number has been ACKed.Thus, when a packet is received in out of order, resend the ACK Thus, when a packet is received in out of order, resend the ACK sent last time -- sent last time -- a duplicate ACKa duplicate ACK!!

60

20

1.0 2.0 3.0 4.0 5.0 6.0 7.0

Time (seconds)

70

304050

10


6

Duplicate ACKs

When a duplicate ACK is seen by the When a duplicate ACK is seen by the sender, it infers that the other side sender, it infers that the other side must have received a packet out of must have received a packet out of order.order.• Delays on different paths could be Delays on different paths could be

different -- thus, the missing packets may different -- thus, the missing packets may be delivered.be delivered.

• So wait for “some” number of duplicate So wait for “some” number of duplicate ACKs before resending data.ACKs before resending data.

• This number is usually 3.This number is usually 3.

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


7

Fast Recovery

When the fast retransmit mechanism signals When the fast retransmit mechanism signals congestion, the sender, instead of returning to Slow congestion, the sender, instead of returning to Slow Start uses a pure AIMD.Start uses a pure AIMD.• Simply reduces the congestion window by half and resumes Simply reduces the congestion window by half and resumes

additive increase.additive increase.

Thus, recovery is faster -- this is called Fast Thus, recovery is faster -- this is called Fast Recovery.Recovery.


8

TCP Vegas

Idea: infer problems from RTT delayIdea: infer problems from RTT delay• Reduce rate before you have lossReduce rate before you have loss

What is a “sign” of congestion:What is a “sign” of congestion:• When RTT increases above a thresholdWhen RTT increases above a threshold• Sending rate flattensSending rate flattens

Decrease sending rate linearlyDecrease sending rate linearlyIssues:Issues:• Estimate RTTEstimate RTT• Set appropriate thresholdSet appropriate threshold

Intuition

Driving on Ice

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

Se

nd

ing

KB

ps 1100

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 8.0

Qu

eu

e s

ize

in r

ou

ter

5

10

2.0 2.5 3.0 3.5 5.0 5.5 6.0 7.0 7.5 8.5

Congestion Window

Average send rate at source

Average Q length in router


10

A Visit to Vegas

Having routers participate in congestion control Having routers participate in congestion control requires changes to core routers -difficult.requires changes to core routers -difficult.It is better to do this end-to-end.It is better to do this end-to-end.However, we want to still have source based control However, we want to still have source based control -- now, it would be source-based congestion -- now, it would be source-based congestion avoidance.avoidance.We need a TCP that watches out for signs of We need a TCP that watches out for signs of congestion --congestion --TCP VegasTCP Vegas..


11

Noting RTT variations

How much does the RTT increase with each packet sent ?How much does the RTT increase with each packet sent ?• Note that with each additional packet, we are adding load.Note that with each additional packet, we are adding load.

One way -- compute for every two round trip delays (with an One way -- compute for every two round trip delays (with an increase in a segment) to see if observed RTT > avg of min increase in a segment) to see if observed RTT > avg of min and maximum RTT.and maximum RTT.If yes, reduce congestion window. If yes, reduce congestion window.


12

A second possibility

Every RTT increase congestion window by a packet (or Every RTT increase congestion window by a packet (or segment).segment).Compute throughput as number of outstanding bytes Compute throughput as number of outstanding bytes divided by RTT.divided by RTT.Also keep the value of the throughput that was achieved Also keep the value of the throughput that was achieved when only one packet was in transit (at the beginning of when only one packet was in transit (at the beginning of the connection).the connection).If the difference between current throughput and this If the difference between current throughput and this above tagged throughput is less than 1/2 decrease above tagged throughput is less than 1/2 decrease congestion window by 1.congestion window by 1.


13

TCP Vegas

Somewhat in line with what we saw in the previous examples.Somewhat in line with what we saw in the previous examples.

Source tries to match the available bandwidth exactly.Source tries to match the available bandwidth exactly.

TCP source maintains what is known as BaseRTT -- the RTT when TCP source maintains what is known as BaseRTT -- the RTT when the flow is not congested -- typically the minimum of all RTTs the flow is not congested -- typically the minimum of all RTTs observed.observed.

It uses this value to determine if or not congestion is being It uses this value to determine if or not congestion is being experienced.experienced.


14

The TCP Vegas: Algorithm

Let BaseRTT be the minimum of all measured RTTs (commonly the Let BaseRTT be the minimum of all measured RTTs (commonly the RTT of the first packet)RTT of the first packet)If not overflowing the connection, thenIf not overflowing the connection, then• ExpectedRate = CongestionWindow / BaseRTTExpectedRate = CongestionWindow / BaseRTT

Source calculates current sending rate (ActualRate) once per RTTSource calculates current sending rate (ActualRate) once per RTTSource compares ActualRate with ExpectedRateSource compares ActualRate with ExpectedRate• Diff = ExpectedRate – ActualRateDiff = ExpectedRate – ActualRate• if Diff < if Diff <

-->increase CongestionWindow linearly -->increase CongestionWindow linearly • else if Diff >else if Diff >

-->decrease CongestionWindow linearly -->decrease CongestionWindow linearly • elseelse

-->leave CongestionWindow unchanged-->leave CongestionWindow unchanged


15

Vegas Behavior

Green Line -- Expected throughputGreen Line -- Expected throughput

Black Line -- Actual throughputBlack Line -- Actual throughput

Shaded area -- region between Shaded area -- region between and and thresholds thresholds

70605040302010

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

2402001601208040

Time (seconds)


16

ParametersParameters : 1 packet: 1 packet• : 3 packets: 3 packets

Even faster retransmitEven faster retransmit• keep fine-grained timestamps for each packetkeep fine-grained timestamps for each packet• check for timeout on first duplicate ACKcheck for timeout on first duplicate ACK

Vegas Parameters


17

Principles of Congestion Control

Congestion:Congestion:informally: “too many sources sending too much informally: “too many sources sending too much data too fast for data too fast for networknetwork to handle” to handle”

different from flow control!different from flow control!

manifestations:manifestations:• lost packets (buffer overflow at routers)lost packets (buffer overflow at routers)• long delays (queueing in router buffers)long delays (queueing in router buffers)

Major research issueMajor research issue


18

Consequences of Congestion

Large delays: throughput vs delay trade-offLarge delays: throughput vs delay trade-off• We don’t want to operate near capacityWe don’t want to operate near capacity

Finite buffers: lost packetsFinite buffers: lost packets

Resending of packets causesResending of packets causes• More packets for the same goodputMore packets for the same goodput• Wasted bandwidth of the packet that gets droppedWasted bandwidth of the packet that gets dropped


19

Causes/costs of congestion: scenario 1

two senders, two two senders, two receiversreceivers

one router, one router, infinite buffers infinite buffers

no no retransmissionretransmission

large delays when large delays when congestedcongested

maximum maximum achievable achievable throughputthroughput


20


one router, one router, finitefinite buffers buffers

sender retransmission of lost packetsender retransmission of lost packet


21


Always: (goodput)Always: (goodput)

If packets are dropped:If packets are dropped:

in

out

=

’in

out

>


22

Causes/costs of congestion: scenario 3 Four senders, multihop paths, timeout/retransmitFour senders, multihop paths, timeout/retransmitCongestion in one link -> retransmits -> congestion Congestion in one link -> retransmits -> congestion in other linksin other links


23


Another “cost” of congestion: when packet dropped, any “upstream transmission capacity used for that packet was wasted!


24

Approaches towards congestion control

End-end congestion End-end congestion control:control:no explicit feedback from no explicit feedback from networknetwork

congestion inferred from congestion inferred from end-system observed loss, end-system observed loss, delaydelay

approach taken by TCPapproach taken by TCP

Network-assisted Network-assisted congestion control:congestion control:routers provide feedback to routers provide feedback to end systemsend systems• single bit indicating single bit indicating

congestion (SNA, DECbit, congestion (SNA, DECbit, TCP/IP ECN, ATM)TCP/IP ECN, ATM)

• explicit rate sender should explicit rate sender should send atsend at

Two broad approaches towards congestion control:


25

Router Assisted Congestion Control

Random Early DetectionRandom Early Detection

Explicit Congestion NotificationExplicit Congestion Notification

Note: often this is referred to as Active Networking: ie Note: often this is referred to as Active Networking: ie routers are involved in perfomance.routers are involved in perfomance.

Active Nets is a much more general ideaActive Nets is a much more general idea


26

RED: Random Early Detection

Idea: routers start dropping packets before Idea: routers start dropping packets before they are congestedthey are congestedBenefits: make behavior smootherBenefits: make behavior smootherHow:How:• When queue is above a When queue is above a thres-1thres-1: drop packets with : drop packets with

probability probability pp

Issues: Issues: • setting the parameterssetting the parameters• Estimating the queue sizeEstimating the queue size


27

• two queue length thresholdstwo queue length thresholds if AvgLen if AvgLen MinThreshold then MinThreshold then

• enqueue the packet

if MinThreshold < AvgLen < MaxThresholdif MinThreshold < AvgLen < MaxThreshold• calculate probability P• drop arriving packet with probability P

if MaxThreshold if MaxThreshold AvgLen AvgLen• drop arriving packet

Thresholds


28

RED: probability P

Not fixedNot fixedFunction of AvgLen and how long since Function of AvgLen and how long since last drop (count) keeps track of new last drop (count) keeps track of new packets that have been queued while packets that have been queued while AvgLen has been between the two AvgLen has been between the two thresholds thresholds • TempP = MaxP * (AvgLen - MinThreshold) /(MaxThreshold - TempP = MaxP * (AvgLen - MinThreshold) /(MaxThreshold -

MinThreshold)MinThreshold)• P = TempP/(1 - count * TempP)P = TempP/(1 - count * TempP)

MaxP is often set to 0.02, meaning that the MaxP is often set to 0.02, meaning that the gateway drops 1 out of 50 packets when gateway drops 1 out of 50 packets when queue size is halfway between queue size is halfway between MinThreshold and MaxThresholdMinThreshold and MaxThreshold


29

Comments on RED

Probability of dropping a particular flow's Probability of dropping a particular flow's packet(s) is roughly proportional to the share packet(s) is roughly proportional to the share of the bandwidth that flow is currently gettingof the bandwidth that flow is currently getting

MaxP is typically set to 0.02, meaning that MaxP is typically set to 0.02, meaning that when the average queue size is halfway when the average queue size is halfway between the two thresholds, the gateway between the two thresholds, the gateway drops roughly one out of 50 packets. drops roughly one out of 50 packets.


30

RED: Dropping probability

P(drop)

1.0

MaxP

MinThresh MaxThresh

AvgLen


31

Selecting Parameters

if traffic is bursty, then MinThreshold if traffic is bursty, then MinThreshold should be sufficiently large to allow link should be sufficiently large to allow link utilization to be maintained at an utilization to be maintained at an acceptably high levelacceptably high levelThe difference between two thresholds The difference between two thresholds should be larger than the typical increase should be larger than the typical increase in the calculated average queue length in in the calculated average queue length in one RTT; setting MaxThreshold to twice one RTT; setting MaxThreshold to twice MinThreshold is reasonable for traffic on MinThreshold is reasonable for traffic on today's Internettoday's Internet


32

Explicit Congestion Notification

Dropping packets = Warn of congestionDropping packets = Warn of congestion

Idea: mark packets to notify congestionIdea: mark packets to notify congestion

How:How:• Congested router marks packet (sets a bit)Congested router marks packet (sets a bit)• Receiver “copies” bit in the ACKReceiver “copies” bit in the ACK• Sender reduces its windowSender reduces its window

Benefit: proactive without losing packetsBenefit: proactive without losing packets

Problem: sender can ignore itProblem: sender can ignore it


33

Current Beliefs

RED + ECN are considered to be goodRED + ECN are considered to be good

RED alone has problemsRED alone has problems


34

Summary

principles behind transport principles behind transport layer services:layer services:• multiplexing/demultiplexingmultiplexing/demultiplexing• reliable data transferreliable data transfer• flow controlflow control• congestion controlcongestion control

instantiation and instantiation and implementation in the Internetimplementation in the Internet• UDPUDP• TCPTCP


35

TCP Flow Control

receiver:receiver: explicitly explicitly informs sender of informs sender of (dynamically (dynamically changing) amount of changing) amount of free buffer space free buffer space • RcvWindow RcvWindow fieldfield in in

TCP segmentTCP segment

sender:sender: keeps the keeps the amount of transmitted, amount of transmitted, unACKed data less unACKed data less than most recently than most recently received received RcvWindowRcvWindow

sender won’t overrunreceiver’s buffers by

transmitting too much, too fast

flow control

receiver buffering

RcvBuffer = size or TCP Receive Buffer

RcvWindow = amount of spare room in Buffer


41

TCP Vegas Details

Value of throughput with no congestion is compared Value of throughput with no congestion is compared to current throughputto current throughputIf current difference is small, increase window size If current difference is small, increase window size linearlylinearlyIf current difference is large, decrease window size If current difference is large, decrease window size linearlylinearlyThe change in the Slow Start Mechanism consists of The change in the Slow Start Mechanism consists of doubling the window every other RTT, rather than doubling the window every other RTT, rather than every RTT and of using a boundary in the difference every RTT and of using a boundary in the difference between throughputs to exit the Slow Start phase, between throughputs to exit the Slow Start phase, rather than a window size value. rather than a window size value.


42

Expected and Actual Rates

The source computes an Expected Rate -- Expected Rate = The source computes an Expected Rate -- Expected Rate = CongestionWindow/BaseRTTCongestionWindow/BaseRTT

Actual Rate is also computed -- number of bytes transmitted Actual Rate is also computed -- number of bytes transmitted during the RTT of a tagged packet.during the RTT of a tagged packet.

Diff = Expected Rate - Actual Rate.Diff = Expected Rate - Actual Rate.

Note that the Diff is always greater than or equal to zero Note that the Diff is always greater than or equal to zero (BaseRTT is the lowest !). (BaseRTT is the lowest !).


43

Vegas Rules

Define two thresholds Define two thresholds and and ..

If Diff < If Diff < , TCP vegas increases congestion window linearly , TCP vegas increases congestion window linearly during next RTT.during next RTT.

If Diff > If Diff > , it decreases the congestion window linearly., it decreases the congestion window linearly.

If If < Diff < < Diff < , TCP vegas leaves the congestion window as , TCP vegas leaves the congestion window as is. is.

advanced networks 20021 transport layer - ii michalis faloutsos many slides from kurose-ross...

Documents

advanced networks

congestion window threshold

congestion control slide

congestion window average

congestion window equal

sign of congestion

time slide

router slide