reliable packet transport - mitweb.mit.edu/6.02/www/f2006/handouts/net_l6.pdfhari balakrishnan, sam...

6.082 Fall 2006 Reliable Packet Transport, Slide 1

Reliable PacketTransport

• “Best Efforts” network layer → lost packets• using ACKnowledgements• Timeout and retransmission• Optimizing performance


Abstraction Layers

Link Layer:Channel access, how bits are transmitted on the channel,framing of data, channel-specific addressing, channel coding.

Network Layer (eg, IP):Addressing and routing fixed-length datagrams through amulti-hop network, breaking up larger packets

Transport Layer (eg, TCP or UDP):Using packets to create illusion of continuous stream ofdata. In-order, connection-based vs. connectionless

Application Layer:HTTP, Mail, FTP, Telnet, RPC, DNS, NFS


“Best Efforts” Network Layer• Packets lead a rough life – they can be lost by the

network for any number of reasons– No current route to destination

• Before transmission is initiated• After packet enters network

– Queue overflow on communications links– Loss of framing on communication channel– Uncorrectable errors detected by CRC– Queue overflow at receiving node

• Rather than tackle each problem separately, adopt an end-to-end approach– Use a single remedy for lost packets, regardless of how

they got lost– Implement in sender and receiver, not a concern for

intermediate nodes


• Network layer provides best-effort service– Loss, delay, variable latency, duplicates, reordering, …– Not convenient for applications

• Transport layer builds on the best effort service to provide applications with a convenient environment– Reliability:

• at least once• at most once

– Performance:• flow and congestion control

– Ordering– Data integrity (checksum)– Timeliness (remove variable latency)

• Also transport provides multiplexing between multiple applications

Transport Layer

Slides are from lectures by Nick Mckeown, Ion Stoica, Frans Kaashoek, Hari Balakrishnan, Sam Madden, and Dina Katabi


At Least Once

• Sender persistently sends until it receives an ack• How long should the timeout be?

– Fixed is bad. RTT changes depending on congestion• Pick a values that’s too big and it will wait too long to

retransmit a packet, • Pick a value too small, and it will unnecessarily retransmit

packets.– Adapt the estimate of RTT adaptive timeout

Timeout and Retransmission

an RTT

Host A Host BData 1

Data 2

ACK

Host A Host BData 1

Data 1

X

RTT = round trip time


RTT Measurements(collected by Caida)


At Most Once• Suppress duplicates• Packets must have ids to allow the receiver to

distinguish a duplicate from a new packet• Receiver should keep track of which packet ids

have been delivered to applications– Discard arriving packets that have already been delivered

(duplicate packets happen when ACKs get lost and sender retransmits not knowing the packet has been delivered).

• To simplify tracking, senders pick monotonically increasing packet ids, i.e., sequence numbers

• Receiver delivers packets to application in order. It keeps track of the largest id delivered so far


How fast should the sender send?

• Waiting for acks is too slow• Throughput is one packet/RTT

– Say packet is 500B – RTT 100ms – Throughput = 40Kb/s, Awful!

• Overlap transmission of new data packets with acks from previous data packets

Host A Host BData 1

Data 2

ACK


Send a window of packets

• Assume the receiver is the bottleneck– Maybe because the

receiver is a slow machine

• Receiver needs to tell the sender when and how much it can send

• The window advances once all previous packets are ackedtoo slow

Host A Host B

Send?

OK, 3 pkts

Idle

2-4

5-7


Sliding Window• Senders advances the

window whenever it receives an ack sliding window

• But what is the right value for the window?

• Note that– if W/RTT < Bottleneck

Capacity under utilization

– If W/RTT > Bottleneck Capacity Large queues

Host A Host B

Send?

OK, 3 pkts

Idle

2-4

3-5


Src

Rcvr

window = 2-6

1 2 3 4 5

p1

a2

x

6

a2

p3

Example


Src

window = 2-6

1 2 3 4 5

p1

a2

x

6

a2

p3

a2

p4

a2

p6

a2

p5

Timeout

2

a7

p2Rcvr

Example


Src

window = 7-11

1 2 3 4 5

p1

a2

x

6

a2

p3

a2

p4

a2

p6

a2

p5

Timeout

2

a7

p2

7 8 9 10 11

Rcvr

In this example, the receiver sent cumulative acks, but the same behavior happens if the receiver acks the received sequence number

Example


What is the right window size?• The window limits how fast the sender sends• Two mechanisms control the window:

– Flow control – Congestion control

• Flow control:– The receiver may be slow in processing the packets

receiver is a bottleneck– To prevent the sender form overwhelming the receiver,

the receiver tells the sender the maximum number of packets it can buffer: fwnd

– Sender sets W ≤ fwnd

But, what if the bottleneck is a slow link inside the network Need Congestion Control


Sharing the Internet

How do you manage the resources in a huge system like the Internet, where users with different interests share the same resources?

Difficult because of:• Size

– Millions of users, links, routers• Heterogeneity

– bandwidth: 9.6kb/s (then modem, now cellular), 10 Tb/s – latency: 50us (LAN), 133ms (wired), 1s (satellite), 260s (Mars)


Congestion

S1

S2

R1 D10Mb/s

2Mb/s

100Mb/sS1

S2

Sources share links,and buffer space

Why a problem?Sources are unaware of current state of resourceSources are unaware of each other

Manifestations:Lost packets (buffer overflow at routers)Long delays (queuing in router buffers)


Congestion CollapseIncrease in input traffic leads to decrease in useful work

Input traffic

Thro

ughp

ut cliffCongestion Collapse

knee

Latency

Causes of Congestion CollapseSpurious retransmissions of packets that are still in flightPacket consume resources and then they are dropped downstream


How do senders learn of congestion?

Potential options:• Router sends a Source Quench to the sender • Router flags the packets indicating congestion• Router drops packets when congestion occurs

– Sender learns about the drop because it notices the lack of ack

– Drops are the solution currently used in the Internet


How do senders learn how much to send?

• Define a congestion control window cwnd• Sender’s window is set to W = min (fwnd, cwnd)• Simple heuristic to find cwnd:

– Sender increases its cwnd slowly until it sees a drop– Upon a drop, sender decreases its cwnd quickly to react

to congestion– Sender increases again slowly

• TCP: Additive Increase / Multiplicative Decrease– Every RTT: No drop: cwnd = cwnd + 1

drop: cwnd = cwnd /2


Additive Increase

D A

Src

Rcvr

cwnd = 1cwnd += 1cwnd = 2

D D A A

cwnd = 3

D D A AD A

cwnd = 4


TCP AIMD

Time

Cwnd

Grab capacity again

Halve Cwnd

Timeout because of a packet loss

desired cwnd

Need the queue to absorb these saw-tooth oscillations


TCP “Slow Start”

D A D D A A D D

A A

D

A

Src

Rcvr

D

A

1 2 4 8

A A A A

How to set the initial cwnd?At the beginning of a connection, increase exponentially

Every RTT, double cwnd


Slow Start + AIMD

Time

Cwnd

Slow start

Additive increase

Multiplicative decrease

Timeout


Summary• It’s possible to build a reliable transport protocol

on top of an unreliable delivery service• Sender and receiver cooperate in ACK/retransmit

scheme– It’s end-to-end since intermediate nodes and links don’t

participate in scheme except for delivering packets just as they’ve always done.

• Want control parameters to adapt to changing network conditions– If parameters are too conservative, network capacity

goes unused– If parameters are too aggressive, problem cascades and

network performance degrades dramatically

reliable packet transport - mitweb.mit.edu/6.02/www/f2006/handouts/net_l6.pdfhari balakrishnan, sam...

Documents