Error and Flow Control - York University · acknowledge data, i.e. both parties implement flow control • piggybacking method: outstanding ACKs are placed in the. header of information

Chapter 5 Peer-to-Peer Protocols

and Data Link Layer Error and Flow Control

CSE 3213, Winter 2010Instructor: Foroohar Foroozan

2

Error ControlApproaches

(2) Error Detection + Automatic Retransmission Req. (ARQ)• not enough redundant info to enable error correction

case (a) receiver detects no errorsan ACK packet is sent back to sender

case (b) receiver detects errorsno ACK sent back to sendersender retransmits frame after a ‘time-out’

(1) Forward Error Correction (FEC)

Error Control

frame

ACK

frame

3

Challenges of ARQ-based Error Control• send one frame at the time, wait for ACK

easy to implement, but inefficient in terms of channel usage

• send multiple frames at once

better channel usage, but more complex to implement -sender must keep (all) sent but unACKed frame(s) in a buffer, as such frame(s) may have to be retransmitted

frame

ACK

frame

frame

ACK

framebuffer of finite size

How many frames should be sentat any point in time?

How should frames be released fromthe sending buffer?

Error Control (Cont.)

4

Flow Control– set of procedures used to restrict the amount of data that

sender can send while waiting for acknowledgment

• two main strategies (1) Stop-and-Wait: sender waits until it receives ACK

before sending next frame(2) Sliding Window: sender can send W frames before

waiting for ACKs

Error Detection + ARQ (error detection with retransmissions)must be combined with methods that intelligently limit the number of

‘outstanding’ (unACKed) frames.

Fewer unACKed frames ⇒ fewer packets buffered at sender and receiver.

Flow and Error Control

Chapter 5Peer-to-Peer Protocols

and Data Link Layer

ARQ Protocols and Reliable Data Transfer

Purpose: to ensure a sequence of information packets is delivered in order and without errors or duplications despite transmission errors & lossesWe will look at:

Stop-and-Wait ARQGo-Back N ARQSelective Repeat ARQ

Basic elements of ARQ:Error-detecting code with high error coverageACKs (positive acknowledgmentsNAKs (negative acknowlegments)Timeout mechanism

Automatic Repeat Request (ARQ)

CRCInformation

packet

Header

Information frame Control frame: ACKs

CRCHeader

Packet Error-freepacketInformation frame

Control frame

Transmitter(Process A)

Receiver(Process B)

Stop-and-Wait ARQ

Timer set after each frame

transmission

• sender sends an information frame to receiver • sender, then, stops and waits for an ACK• if no ACK arrives within time-out, sender resends the frame, and again stops

and waits time-out period > roundtrip time

8

Lost Acknowledgment• frame is received correctly, but ACK undergoes

errors / lossafter time-out period, sender resends framereceiver receives the same frame twice

• frames must be numbered so that receiver canrecognize and discard duplicate frames

sequence number are included in packet header

ACK

How will receiver knowthat this is NOTa new packet?!

ACK

ACK

retransmittedframe

ACK

0

ACK

ACK

Receiver has alreadyreceived frame 2 –

it resends an ACK and discards the duplicate.

1

2

2

without packet numbering with packet numbering

frame 0

frame 1

frame 2

frame 2

frame

frame

frame

frame

• abnormalities:(1) lost acknowledgment(2) delayed acknowledgment

Stop-and-Wait ARQ (Cont.)

9

ACKreceiver sees this asACK for 2nd frame-0

and sends frame-1

Delayed Acknowledgment(Premature Timeout)

• ACKs can be delayed due to problems withlinks or network congestion

time-out expires early, sender resends frame

when delayed ACK arrives, sender assumesthat given ACK corresponds to last frame sent

• ACKs must be numbered to prevent gaps indelivered packet sequence

0

0

1receiver sees this asACK for frame-1

and sends frame-2

ACK

2

frame-1 not delivered !!!

ACK 1

0

0

1 ACK 1

cannot send frame-2

0

0

How large should the packet / ACK sequence be? Only 1-bit long !!!without ACK numbering with ACK numbering

frame 0

frame 0

frame 0

frame 0


10


frametf time

A

B

tprop tacktproc tprop

tproc

t0 = total time to transmit 1 frame

Stop-and-Wait Model

Rn

Rn

tt

ttttt

afprocprop

ackfprocprop

+++=

+++=

22

220 bits/info frame

channel transmission rate

bits/ACK frame

S&W Efficiency on Error-free channel

.)(21

10

0

f

procprop

f

a

f

oof

eff

nRtt

nn

nn

Rt

nn

RR

+++

−=

−

==η

bits for header & CRC

,bitsn informatio edeliver th torequired timetotal

ndestinatio todelivered bitsn informatio ofnumber

0

0

tnn

R ofeff

−==

Effect offrame overhead

Effect ofACK frame

Effect ofDelay-Bandwidth Product

Effective transmission rate:

Transmission efficiency:

13

Bandwidth-delay product = 2*(tprop + tproc)*R = = capacity of the transmission pipe from the sender to the receiver and back.


14

frame size in comparisonto bandwidth-delay product

max number of bits in transit –‘the pipe is full’

tprop + tproc > tframe tprop + tproc < tframe

Stop-and-Wait ARQ becomes inadequate when data is fragmented into small frames, such that nf / R = tframe is small relative to tprop .


15

Example [ impact of delay-bandwidth product ]

nf = 1250 bytes = 10000 bitsnACK = nheader = 25 bytes = 200 bits

Efficiency 200 km

(tprop = 1 ms)2000 km

(tprop = 10 ms)20000 km

(tprop = 100 ms)200000 km

(tprop = 1 sec)

1 Mbps103

88%104

49%105

9%106

1%

1 Gbps106

1%107

0.1%108

0.01%109

0.001%

Stop-and-Wait does NOT work well for very high speeds or long propagation delays.

0.02n

nn

n

f

header

f

ACK ==⇒

f

procprop

f

procprop

f

ACK

f

header

effSW

n)Rt(t2

1.02

0.98

n)Rt(t2

nn1

nn1

RR

+⋅+

=+⋅

++

−==η


16

• Pf = probability that transmitted frame has errors and need to be retransmitted

(1-Pf) – probability of successful transmission

– average # of (re)transmission until first correct arrival

total delay per frame:

Stop-and-WaitEfficiency inChannel with Errors

fP1-1

f00 P1-

1tretrans.) of #(averaget ⋅=⋅

f

procprop

f

ACK

f

header

ff

0

headerf

eff_errorSW_error

n)Rt2(t

nn1

nn1

P(1-RP(1-

tnn

RR

+++

−⋅=

−

== ))η

0fSW_error P(1- ηη ⋅= )

Pf increases ⇒ ηSW decreases

(∗)

and including


170

ff

f00 t

P1-1

P1-Pout-timet]E[out-timet error in transmiss of# ≈⋅+=⋅+

successfultransmission

Probability that i transmission are needed to deliver frame successfully( i-1 transmission in error and the ith transmission is error free ):

P[ # of trans. in error = i-1 ] = (1-Pf) Pfi-1

Total averagedelay per frame:

f

f

2f

ff1n

1nfff

1n

nff

1i

1iff

1i

1iff

1ierror in trans

P1P

)P(11P)P(1PnP)P(1

Pn)P(1P1)-(i)P(1

)PP(11)-(i1]-iP[n1)-(ierror] in onstransmissi of E[#

−=

=−

⋅⋅−=⋅⋅⋅−=

=⋅⋅−=⋅⋅−=

=−⋅==⋅=

∑

∑∑

∑∑

∞

=

−

∞

=

∞

=

−

∞

=

−∞

=

time-out time-out time-out t0

average # of transmissions in errorbefore a successful transmission


18

Piggybacking • Stop-and-Wait discussed so far was ‘unidirectional’

• in ‘bidirectional’ communications, both parties send andacknowledge data, i.e. both parties implement flow control

• piggybacking method: outstanding ACKs are placed in theheader of information frames

• piggybacking can save bandwidth since the overhead froma data frame and an ACK frame (addresses, CRC, etc) canbe combined into just one frame

S(0)

0

R(0)S(1)

01

R(1)S(2)

1

S(0)

R(0)

S(1)

R(1)

S(2)

without piggybacking with piggybacking


Applications of Stop-and-Wait ARQ

IBM Binary Synchronous Communications protocol (Bisync): character-oriented data link controlXmodem: modem file transfer protocolTrivial File Transfer Protocol (RFC 1350): simple protocol for file transfer over UDP

Go-Back-N ARQImprove Stop-and-Wait by not waiting!Keep channel busy by continuing to send framesAllow a window of up to Ws outstanding framesUse m-bit sequence numberingIf ACK for oldest frame arrives before window is exhausted, we can continue transmittingIf window is exhausted, pull back and retransmit all outstanding framesAlternative: Use timeout

21

Go-Back-N ARQ – overcomes inefficiency of Stop-and-Wait ARQ –sender continues sending enough frames to keepchannel busy while waiting for ACKs

• a window of Ws outstanding frames is allowed• m-bit sequence numbers are used for both - frames

and ACKs, and Ws = 2m-1

Assume: Ws= 41) sender sends frames one by one2) frame 3 undergoes transmission error – receiver ignores frame 3 and all subsequent frames3) sender eventually reaches max number of outstanding frames, and takes following action:

go back N=Ws frames and retransmit all frames from 3 onwards

Go-Back-N ARQ

22

Sender Sliding Window • all frames are stored in a buffer, outstandingframes are enclosed in a window

frames to the left of the window are already ACKedand can be purged

frames to the right of the window cannot be sentuntil the window slides over them

whenever a new ACK arrives, the window slidesto include new unsent frames

once the window gets full (max # of outstandingframes is reached), entire window gets resent

Receiver Sliding Window

before ACKs for frames 0 and 1 arrive

after ACKs for frames 0 and 1 arriveand window slides

• the size of receiver window is always 1

receiver is always looking for a specific frameto arrive in a specific order

any frame arriving out of order is discardedand needs to be resent

The complexity of the receiver in Go-Back-N is the same as that of Stop-and-Wait!!!Only the complexity of the transmitter increases.

Go-Back-N ARQ (Cont.)

23

Problems with Go-Back-N(Go-Back-N with Timeout)

• Go-Back-N works correctly (retransmission ofdamaged frames gets triggered) as long as thesender has an unlimited supply of packets that need to be transmitted

but, in case when packets arrive sporadically, there may not be Ws-1 subsequent transmissions⇒ window will not be exhausted, retransmissionswill not be triggered

this problem can be resolved by modifyingGo-Back-N such that:1) set a timer for each sent frame2) resend all outstanding frames either when

the window gets full or when the timer offirst frame expires


24

Example [ lost frame in Go-Back-N with time-out ]

Note: • ACKs number always defines the number of the next expected frame !!!• in Go-Back-N, receiver does not have to acknowledge each frame received –

it can send one cumulative ACK for several frames


25

Sequence Numbers and Window Size • m bits allotted within a header for sequence numbers

⇒ 2m possible sequence numbers

how big should the sender window be!?

W > 2m cannot be accepted – multiple frames withsame seq. number in the window ⇒ ambiguous ACKs

W = 2m can still cause some ambiguity – see below

W = 2m – 1 acceptable !!!

ACK1

ACK2

ACK3

ACK0

ACK1

ACK1

ACK2

ACK3

ACK3

window size 2m = 4 window size 2m-1 = 3


26

• completely efficient if Ws is large enough to keep channel busy,and if channel is error free

• in case of error-prone channel, with Pf frame loss probability,time to deliver a frame is:

- if 1st transmission succeeds – prob. (1-Pf)

- if 1st transmission does NOT succeeds –prob. Pf

• total average time required to transmit a frame:

• transmission efficiency

Go-Back-NEfficiency

framesf

frame tWP1

1t ⋅⋅−

+

frametaverage # of

frame/window (re)transmission

until a successful transmission

framesf

fframes

fframefframefGBN tW

P1PtW

P11tPt)P(1t ⋅⋅

−+=⎟⎟

⎠

⎞⎜⎜⎝

⎛⋅

−+⋅+⋅−=

)P(11)P(W1

nn1

Rtnn

ffs

f

header

GBN

headerf

GBN −−+

−=

−

=η (∗∗)


27

0 1 2 3 4 5 6 7 8 9 1011

f

f

P1Perror] in onstransmissi of E[#−

=

frameGBN tt =

frameSf

fframeGBN tW

P1Ptt ⋅−

+=

What is total average time required to transmit a frame, assuming Pf?successful

transmission

0 1 2 3 4 5 6 7 R R R R R R R R 1110

WS WS

8 9


tframe

1st attempt successful:

frameSframeGBN tWtt ⋅+=2nd attempt successful:

frameSframeGBN tWerror] in onstransmissi of E[#tt ⋅⋅+=average case:

)P(11)P(W1

nn1

Rtnn

ffs

f

header

GBN

headerf

GBN −−+

−=

−

=η⇒

tframeACK keeps window ‘sliding’


28

Example [ Stop-and-Wait vs. Go-Back-N ]

nf = 1250 bytes = 10000 bitsnACK = nheader = 25 bytes = 200 bits

Compare S&W with GBN efficiency for random bit errors with pb = 0, 10-6, 10-5, 10-4 and bandwidth-delay product R*2*(tprop+tproc) = 1 Mbps * 100 ms = 100000 bits = 10 frames → use Ws = 11.

Efficiency pb=0 pb=10-6 pb=10-5 pb=10-4

S&W 8.9% 8.8% 8.0% 3.3%GBN 98% 88.2% 45.4% 4.9%

• Go-Back-N provides significant improvement over Stop-and-Wait for large delay-bandwidth product

• Go-Back-N becomes inefficient as error rate increases


Applications of Go-Back-N ARQ

HDLC (High-Level Data Link Control): bit-oriented data link controlV.42 modem: error control over telephone modem links

Selective Repeat ARQGo-Back-N ARQ inefficient because multiple frames are resent when errors or losses occurSelective Repeat retransmits only an individual frame

Timeout causes individual corresponding frame to be resentNAK causes retransmission of oldest un-acked frame

Receiver maintains a receive window of sequence numbers that can be accepted

Error-free, but out-of-sequence frames with sequence numbers within the receive window are buffered Arrival of frame with Rnext causes window to slide forward by 1 or more

31

Selective Repeat ARQ • Go-Back-N is NOT suitable for ‘noisy links’ – incase of a lost/damaged frame a whole windowof frames need to be resent

excessive retransmissions use up the bandwidthand slow down transmission

• Selective Repeat ARQ overcomes the limitationsof Go-Back-N by adding 2 new features

(1) receiver window > 1 frame, so that out-of-orderbut error-free frames can be accepted

(2) retransmission mechanism is modified – only individual frames are retransmitted

• Selective Repeat ARQ is used in TCP !!!

sender window of size WS receiver window of size WR

Selective Repeat ARQ

32

Receiver:• window advances whenever next

in-order frame arrives• out-of-order frames are accepted only

if their sequence numbers satisfy

Rnext < Rframe < Rnext + Ws

• a negative ACK (NAK) with sequencenumber Rnext is sent whenever anout-of-sequence frame is observed

Sender:• window advances whenever an ACK

arrives• if a timer expires, the corresponding

frame is resent, and the timer is reset

• whenever a NAK arrives, Rnext frameis resent

Selective Repeat ARQ Operation

Rnext

Rnext + WS -1


33

Window Sizes – WS and WR • m bits allotted within a header for sequencenumbers ⇒ 2m possible sequence numbers

how big should the windows be!?

WS and WR = 2m-1 cannot be accepted dueto possible ambiguity as shown below

W = 2m/2 = 2m-1 acceptable !!!

ACK1

ACK2

ACK3

ACK1

ACK2

window size 2m-1 = 3 window size 2m-1 = 2


34

Selective RepeatEfficiency

• completely efficient if Ws is large enough to keep channelbusy, and if channel is error free

of course, sequence number space must be 2X sequencesequence number space of Go-Back-N

• in case of error-prone channel, total average time requiredto transmit a frame:

• transmission efficiency

)P(1Rn

P1tt

f

f

f

frameSR −⋅

=−

=

)P(1n

n1Rtnn

RRη f

f

headerSR

headerf

effSR −⋅⎟⎟

⎠

⎞⎜⎜⎝

⎛−=

−

== (∗∗∗)


35

0 1 2 3 4 5 6 7 8 9 1011

f

f

P1P−

frameSR tt =

Rn

P11t

P1Ptt f

fframe

f

fframeSR ⋅

−=⋅

−+=

What is total average time required to transmit a frame, assuming Pf?successful

transmission

0 1 2 3 4 5 6R 8 9 1011127

on NAK or time-out


tframe

1st attempt successful:

frameframeSR ttt +=2nd attempt successful:

frameframeSR terror] in onstransmissi of E[#tt ⋅+=average case:

)P(1n

n1Rtnn

ff

headerSR

headerf

SR −⎟⎟⎠

⎞⎜⎜⎝

⎛−=

−

=η⇒

NAK 0


36

PerformanceComparison

• assume nACK and nheader are negligible relative to nf, and

• efficiencies of three ARQ techniques are

• for 0 < Pf < 1, Selective Repeat provides best performance

• for Pf → 0 Go-Back-N as good as Selective Repeat

size of the “pipe” inmultiples of frames

( )fSW P1-L1

1⋅

+=η

)P(1LP11

ff

GBN −+

=η

)P(1 fSR −=η

ηSW < ηGBN < ηSR

1−==+

sf

procprop WLn

)Rt2(t WS is for 1 less than thenumber of frames currently in transit

Stop-&-Wait vs. Go-Back-N vs. Selective Repeat

37

ARQ Efficiency Comparison

0

0.5

1

1.5

-9 -8 -7 -6 -5 -4 -3 -2 -1

- LOG(p)

Effic

ienc

y

SelectiveRepeat

Go Back N 10

Stop and Wait100

Go Back N 100

Stop and Wait10

10-9 10-8 10-7 10-6 10-5 10-4 10-3 10-2 10-1

p

Delay-Bandwidth product = 10, 100

Documents

Error and Flow Control - York University · acknowledge data, i.e. both parties implement flow control • piggybacking method: outstanding ACKs are placed in the. header of information