4

TCP#Congestion#Control
TCP#Tahoe
TCP#Reno
TCP#vegas
!nternet#Congestion#Collapse
!n#the#late#;3s/#the#!nternet#suffered#a#
congestion#collapse
Host
Host
Host
Congestion$
Packet#drops$$
Data .#Retransmissions
Nore#Drops$$$
Collapse
Why#is#Today's#Topic#
!mportant?
7KH#DOJRULWKP#IRU#7&3#FRQJHVWLRQ#
FRQWURO#LV#WKH#PDLQ#UHDVRQ#ZH#FDQ#
XVH#WKH#,QWHUQHW#VXFFHVVIXOO\#WRGD\#
GHVSLWH#ODUJHO\#XQSUHGLFWDEOH#XVHU#
DFFHVV#SDWWHUQV#DQG#GHVSLWH#
UHVRXUFH#ERWWOHQHFNV#DQG#
OLPLWDWLRQV1#:LWKRXW#7&3#FRQJHVWLRQ#
FRQWURO/#WKH#,QWHUQHW#FRXOG#KDYH#
EHFRPH#KLVWRU\ D#ORQJ#WLPH#DJR1
Resource#Nanagement#
Solutions
Handling#congestion
pre0allocate#resources#so#as#to#avoid#congestion#/
control#congestion#if#+and#when,#is#occurs#-
Two#points#of#implementation
routers#inside#the#network#+queuing#discipline,#/
hosts#at#the#edges#of#the#network#+transport#protocol,#-
Destination
1.5-Mbps T1 link
Router
Source
2
Source
1
100-M
bps FDDI
10-Mbps Ethernet
Problem#with#;3s#Transport#
Protocol#+Early#TCP,
The#connection#flow#rate#does#not#depend#
on#the#level#of#congestion#in#the#network
The#;3s#question#was=#How#to#re0design#
TCP#to#send#less#traffic#when#the#network#
gets#congested?
How#to#detect#congestion?
How#to#make#flow#rate#sensitive#to#congestion#
level?
Detecting#Congestion
Packet#drops#indicate#congestion
!s#that#really#true?
Why#does#it#work?
Src
Dst
Packet
Ack
Drop
Timeout$#No#Ack#
=#Congestion$
5
Controlling#Congestion#- The#
Effect#of#Window#Size
Note#that#sender's#window#is#equal#to#the#
number#of#sender#packets#in#flight#+in#the#
network,1#Why?
A#Window's#worth#of#packets
X#acks
Window
X#more#packets
Source####################################Destination
Controlling#Congestion
Reduce#window# less#packets#in#the#
network
!ncrease#window# more#packets#in#the#
network
!dea=#Concept#of#a#FRQJHVWLRQ#ZLQGRZ -
window#is#smaller#when#congestion#is#
larger#and#vice#versa
van#Jacobson's#Congestion#
Control
van#Jacobson#+formerly#C!SCO#Chief#
Technical#Officer#- now#a#startup#
founder,#introduced#congestion#control#
into#TCP#in#4<;;04<;<1
The#original#version#of#TCP#that#
implements#van#Jacobson's#congestion#
control#is#known#as#TCP#Tahoe1#
Elements#of#Congestion#
Control#in#TCP#Tahoe
Additive#increase/#multiplicative#
decrease
Slow#start
Fast#retransmit
Additive#!ncrease/#
Nultiplicative#Decrease
Each#time#a#packet#drop#occurs/#slash#
window#size#in#half#+multiplicative#
decrease,
Nultiplicative#decrease#is#necessary#to#avoid#
congestion
When#no#losses#are#observed/#gradually#
increase#window#size#+additive#increase,
A!ND#+cont,
!n#practice=#increment#a#little#for#each#ACK
Increment = (MSS*MSS)/CongestionWindow
CongestionWindow += Increment
Source Destination

Algorithm
increment#CongestionWindow by#
one#packet#per#RTT#+OLQHDU#LQFUHDVH,
divide#CongestionWindow by#two#
whenever#a#timeout#occurs#
+PXOWLSOLFDWLYH#GHFUHDVH,
6
A!ND#+cont,
Trace=#sawtooth#behavior
60
20
1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0
K
B
Time (seconds)
70
30
40
50
10
10.0
Problems#
What#should#the#window#size#be
!nitially?
Upon#packet#loss#and#timeout?
Pessimistic#window#size?#+e1g1/#4,
Additive#increase#is#too#slow#in#ramping#up#
window#size#- short#connections#will#not#
fully#utilize#available#bandwidth
Optimistic#window#size?
Large#initial#burst#may#cause#router#queue#
overflow#
Slow#Start
Objective=#determine#the#available#
capacity#quickly
!dea=
Use#CongestionThreshold as#an#
optimistic#CongestionWindow estimate
begin#with#CongestionWindow =#4#
packet
double#CongestionWindow each#RTT#
+increment#by#4#packet#for#each#ACK,
When#CongestionThreshold is#
crossed/#use#additive#increase#
Source Destination

Slow#Start#!llustration
Assume#that#
CongestionThreshold = 8
cwnd = 1
cwnd = 2
cwnd = 4
cwnd = 8
cwnd = 9
cwnd = 10
3
5
7
9
;
43
45
47
W

3
W

5
W

7
W

9
Roundtrip times
C
o
n
g
e
s
t
i
o
n

W
i
n
d
o
w

(
i
n

s
e
g
m
e
n
t
s
)
threshold
Figures#by#Jorg Liebeherr
Slow#Start#+cont,
Used.
when#first#starting#connection
when#connection#goes#dead#waiting#for#timeout
Real#TCP#Trace
60
20
1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0
K
B
70
30
40
50
10
Fast#Retransmit
Problem=#coarse0grain#
TCP#timeouts#lead#to#
idle#periods
Fast#retransmit=#
Send#an#ack on#every#
packet#reception
Send#duplicate#of#last#
ack when#a#packet#is#
received#out#of#order
Use#duplicate#ACKs#to#
trigger#retransmission
Packet 1
Packet 2
Packet 3
Packet 4
Packet 5
Packet 6
Retransmit
packet 3
ACK 1
ACK 2
ACK 2
ACK 2
ACK 6
ACK 2
Sender Receiver
7
TCP#Reno#and#Fast#Recovery
Skip#the#slow#start#phase
So#directly#to#half#the#last#successful#
CongestionWindow
60
20
1.0 2.0 3.0 4.0 5.0 6.0 7.0
K
B
70
30
40
50
10
Recap#of#Congestion#Control
TCP#Reno#and#its#derivatives#are#most#widely#
used#today
Performance#issue=#try#to#avoid#forcing#the#source#
to#go#to#slow0start#+Why?,#
Source#goes#to#slow0start#initially#and#upon#
timeouts#+what's#the#rationale#for#that?,
Source#cuts#congestion#window#in#half#upon#a#fast#
retransmit#+why#not#go#to#slow#start?,
Single#packet#drops#can#be#caught#by#the#fast#
retransmit2fast#recovery#algorithm#- source#does#
not#go#to#slow#start#+Why?,
Nultiple#consecutive#packet#drops#will#most#likely#
force#the#source#into#slow#start#+why?,#
Self#Clocking#and#Slow#Start
Each#packet's#transmission#is#clocked"#by#an#
ACK#- no#bursts#develop
W=4####W=5###W=7####W=8####W=9####W=:
Slow#Start
.
Self#Clocking#in#Operation
Each#packet's#transmission#is#clocked"#by#an#
ACK#- no#bursts#develop
W=65
. .
Self#Clocking#!nterrupted
During#timeouts/#ACKs#are#drained#from#the#
network1#Self#clocking#is#interrupted1#Next#
transmission#causes#a#burst1#Hence/#slow#start$
W=65
. .
Lost
Timeout
Retransmission
Ack65
49#packet#burst
Cut#window#in#425
ACKs
Drained$$
Self#Clocking#and#Fast#
Retransmit#2#Fast#Recovery
When#fast#retransmit#is#used/#the#packet#is#
retransmitted#before#all#ACKs#are#drained1#Slow#
start#is#not#needed
W=65
. .
Lost
Fast
Retransmission
Cut#window#in#425
8
Tahoe2Reno#
Retransmission#Timers
The#retransmission#timers#are#set#based#on#round0
trip#time#+RTT,#measurements#that#TCP#performs
Jorg Liebeherr
pkt4
ack5
pkt5#pkt6#pkt7
ack8
pkt8#pkt9#pkt:
ack;
Round0Trip#Time#
Neasurements
TCP#uses#RTT#mean#and#variance#estimators/#called#
VUWW and#UWWYDU=
VUWW
Q.4
# 577##.#+40 ,#VUWW
Q
UWWYDU
Q.4
# +#_#577#0 VUWW
Q.4
_#,#.#+40 ,#UWWYDU
Q
RTO#is#set#to#the#mean#RTT#plus#four#times#the#
estimated#RTT#variance1
572
Q.4
VUWW
Q.4
.#7 UWWYDU
Q.4
Problem
!f#a#segment#is#retransmitted/#TCP#can't#
compute#the#RTT1
Why?
pkt4
pkt4
ack5
Karn's#Algorithm
Don't#update#RTT#on#any#segments#that#
have#been#retransmitted1
When#TCP#retransmits/#set=
572
Q.4
#PD[#+#5#572
Q
/#97,
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
Alternative#strategy
predict#when#congestion#is#about#to#happen
reduce#rate#before#packets#start#being#discarded
called#congestion#DYRLGDQFH
Two#possibilities#
router0centric=#DECbit#and#RED#Gateways#
host0centric=#TCP#vegas#
TCP#vegas
Uses#congestion#avoidance#instead#of#
congestion#control
vegas=#Congestion#avoidance= Predict#and#
avoid#congestion#EHIRUH it#occurs
Tahoe/#Reno=#Congestion#control= React#to#
congestion#DIWHU it#occurs
Question=#How#to#predict#congestion?
9
TCP#vegas
!dea=#source#
watches#for#some#
sign#that#router's#
queue#is#building#
up#and#congestion#
will#happen#too>#
e1g1/
RTT#grows
sending#rate#
flattens
60
20
0.5 1.0 1.5 4.0 4.5 6.5 8.0
K
B
Time (seconds)
Time (seconds)
70
30
40
50
10
2.0 2.5 3.0 3.5 5.0 5.5 6.0 7.0 7.5 8.5
900
300
100
0.5 1.0 1.5 4.0 4.5 6.5 8.0
S
e
n
d
in
g
K
B
p
s
1100
500
700
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
Q
u
e
u
e
s
iz
e
in
ro
u
te
r
5
10
2.0 2.5 3.0 3.5 5.0 5.5 6.0 7.0 7.5 8.5
Observation
Packet#accumulation#in#the#network#can#be#
inferred#by#monitoring#RTT#and#sending#rate#
Bottleneck#Link
Overloaded
Router
S
e
n
d
in
g
#R
a
te
FZQG
Sending#Rate#=#FZQG 2#RTT
Algorithm#
Let#BaseRTT be#the#minimum#of#all#measured#RTTs#
+commonly#the#RTT#of#the#first#packet,
!f#not#overflowing#the#connection/#then
ExpectRate = CongestionWindow/BaseRTT
Source#calculates#ActualRate once#per#RTT
Source#compares#ActualRate with#ExpectRate
Diff = ExpectedRate - ActualRate
if Diff <
increase CongestionWindow linearly
else if Diff >
decrease CongestionWindow linearly
else
leave CongestionWindow unchanged
Algorithm#+cont,
Parameters
=#4#packet
=#6#packets
70
60
50
40
30
20
10
K
B
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
C
A
M
K
B
p
s
240
200
160
120
80
40
Time (seconds)
Congestion#Avoidance#in#the#
Network#Layer=#DECbit
!n#the#network=
Router#computes#average#queue#length
!f#average#>#4/#set#congestion#bit#in#packet#header
Destination#copies#congestion#bit#onto#the#ACK
On#the#source=
Source#maintains#a#sliding#window#+as#in#TCP,1
!f#less#than#83(#of#the#ACKs#in#last#window#have#
congestion#bit#set/#increase#window#by#41
!f#more#than#83(#of#the#ACKs#have#congestion#bit#set/#
shrink#window#to#31;:8#of#its#current#size1#
DECbit
Computing#the#average#queue#length
Average#carried#out#over#last#busy.idle#intervals#
plus#the#current#busy#interval
Queue length
Current
time
Time
Current
cycle
Previous
cycle
Averaging
interval
:
Congestion#Avoidance#in#RED
Source=
Sally#Floyd#and#van#Jacobson/ Random#
Early#Detection#Gateways#for#Congestion#
Avoidance/"#,(((2$&0#7UDQVDFWLRQV#RQ#
1HWZRUNLQJ/ vol1#4/#No1#7/#pp1#6<:0746/#
August#4<<61
Random#Early#Detection#+RED,
Notification#is#implicit#
just#drop#the#packet#+TCP#will#timeout,
could#be#made#explicit#by#marking#the#packet
Early#random#drop
rather#than#wait#for#queue#to#become#full/#drop#each#
arriving#packet#with#some#GURS#SUREDELOLW\#whenever#
the#queue#length#exceeds#some#GURS#OHYHO
RED#Details
Compute#average#queue#length
AvgLen = (1 - Weight) * AvgLen + Weight * SampleLen
3#?#Weight ?#4#+usually#31335,
SampleLen is#queue#length#each#time#a#packet#arrives
MaxThreshold MinThreshold
AvgLen
RED#Details#+cont,
Two#queue#length#thresholds
if AvgLen <= MinThreshold then
enqueue the packet
if MinThreshold < AvgLen < MaxThreshold
then
calculate probability P
drop arriving packet with probability P
if ManThreshold <= AvgLen then
drop arriving packet
RED#Details#+cont,
Computing#probability#P#+first#pass,=
P = MaxP * (AvgLen - MinThreshold)/
(MaxThreshold - MinThreshold)
Drop#Probability#Curve
P(drop)
1.0
MaxP
MinThresh MaxThresh
AvgLen
RED#Details#+cont,
Problem#with=
P = MaxP * (AvgLen - MinThreshold)/
(MaxThreshold - MinThreshold)
Drop#probability#is#insensitive#to#time#since#last#
drop#- two#or#more#drops#may#occur#back#to#back#
forcing#the#source#into#slow#start
!mprovement=
TempP = MaxP * (AvgLen - MinThreshold)/
(MaxThreshold - MinThreshold)
P#=#TempP2+4#- count-TempP,
;
R!O=#RED#with#!n#and#Out
Packets#fall#into#two#categories
!n#profile#+high#priority,
Out#of#profile#+low#priority,1
Out#packets#are#dropped#first