You are on page 1of 15

Congestion Control

Gaurav S. Kasbekar
Dept. of Electrical Engineering
IIT Bombay

1-1
References
 Slides by Kurose and Ross available at:
http://ctas.poly.asu.edu/millard/CET459/lectno/K%
20-%20R%20stuff/index.html
 Bertsekas and Gallager, Chapter 6

1-2
Congestion
 Informally: “too many sources sending too much data
too fast for network to handle”
 Effects:
 long delays (queuing in router buffers)
 lost packets (buffer overflow at routers)
 Reliable data transfer techniques overcome only
effects of congestion, not the cause
 To overcome the cause, congestion must be detected
and sources must reduce transmission rates
 “Congestion Control”

3-3
Outline
 Adverse effects of congestion
 via simple network scenarios

 Approaches to congestion control

3-4
Scenario 1
 Infinite buffer Host A lout
lin : original data
 Capacity of outgoing link of
router: R
 No packet losses, no Host B unlimited shared
output link buffers
retransmissions
 What’s a “good” value of 𝜆𝑖𝑛 ?
 As 𝜆𝑖𝑛 approaches R/2, avg.
thru. increases, but delay
increases

Plot of 𝝀𝒐𝒖𝒕 vs 𝝀𝒊𝒏


Plot of avg. delay vs 𝝀𝒊𝒏
Throughput and Delay
 Both arrival processes have same average
rate (< R/2)
 Avg. delay:
 close to 0 for first process
 as in previous slide for second process
• since arrivals random
• typically the case in practice
Power
 Trade-off between avg. throughput and
delay
 Power = Avg. Thru. / Avg. Delay
 sometimes used to measure network
performance

3-7
Scenario 1: Conclusion
 Large queuing delays are experienced as
packet arrival rates approach link
capacities

3-8
Scenario 2
 Same as scenario 1, except:
 finite buffers
 sender retransmission of lost packets
 Note: 𝜆′𝑖𝑛 ≥ 𝜆𝑖𝑛
 For small enough 𝜆𝑖𝑛 , relation between 𝜆′𝑖𝑛 and 𝜆𝑜𝑢𝑡 :
 𝜆′𝑖𝑛 ≈ 𝜆𝑜𝑢𝑡
 Gap 𝜆′𝑖𝑛 − 𝜆𝑜𝑢𝑡 is measure of wasted work

Host A lin : original lout


data
l'in : original data, plus
retransmitted data

Host B finite shared output


link buffers
Scenario 2: Plot of 𝜆𝑜𝑢𝑡 vs 𝜆′𝑖𝑛
 Case (a): Sender transmits only when free slot
known to be available in buffer
 Case (b): Sender transmits only when packet
known to be lost with certainty
 Case (c): Sender also transmits when packet
excessively delayed but not lost
Scenario 2: Conclusions
 Retransmissions due to
i. dropped packets
ii. packets excessively delayed but not lost

 Need to transmit at rate 𝜆′𝑖𝑛 > 𝜆𝑜𝑢𝑡 to


achieve “goodput” 𝜆𝑜𝑢𝑡
 In case (ii), packets are dropped by
destination
 link capacity used in transporting packets
wasted

3-11
Scenario 3
 Multihop paths
 Outgoing links of routers: bandwidth R
 Incoming links from hosts to routers: high bandwidth

Host A lout
lin : original data
l'in : original data, plus
retransmitted data

finite shared output


link buffers

Host B

3-12
Scenario 3: Plot of 𝜆𝑜𝑢𝑡 vs 𝜆′𝑖𝑛

H l
o
o
s
u
t
A t

H
o
s
t
B

 Decrease in 𝜆𝑜𝑢𝑡 if 𝜆′𝑖𝑛 increases beyond a point

3-13
Scenario 3: Conclusions
 When multihop paths present, any upstream
transmission capacity used for dropped packets is
wasted
 Effect worsens as number of hops increases
Scenario 4

 Consider the n-link user:


 only 1/2 units/ sec of its flow reaches router 2
 only 1/3 units/ sec reaches router 3
 … 1/k units/ sec reaches k’th router

You might also like