Module 7: Transport Layer - Congestion and TCP Variants

Question 1

Solution for congestion in TCP

Answer 1

Dynamic adjustment
- every sender infers the level of congestion
- adapts its sending rate for the greater good

Question 2

Why is congestion control difficulty?

Answer 2

How does the sender know there is congestion?
How should the sender adapt?
What is the performance objective?

Question 3

The goal of congestion control is to…

Answer 3

prevent communicating hosts from overloading the network

Question 4

How does the sender infer congestion?

Answer 4

1. packet delay (RTT estimate)
2. packet loss (timeout, triple duplicate ack)

Question 5

What can the end host do when it detects congestion?

Answer 5

decrease the sending rate

when NOT detecting congestion
- increase sending rate
- make sure packets are getting sent

Question 6

What are the pros and cons of the TCP Congestion Window?

Answer 6

• pro: avoids having explicit feedback from network
• con: under-shooting and over-shooting the rate

Question 7

How does TCP increase and decrease congestion window size?

Answer 7

increase: linearly
decrease: multiplicatively

Question 8

What’s worse? Over-sized window or under-sized window?

Answer 8

Oversized is worse
Oversized: packets are dropped and retransmitted (congestion collapse)
Undersized: somewhat lower throughputs

Question 9

What causes the sawtooth behavior of congestion window size?

Answer 9

additive increase and multiplicative decrease

Question 10

How is congestion window size calculated?

Answer 10

Question 11

What is the difference between flow control and congestion control?

Answer 11

flow control: keep a fast sender from overwhelming a receiver

congestion control: keep a set of senders from overloading the network

Similar mechanisms:
TCP flow control: receiver window
TCP congestion control: congestion window
TCP window: min{congestion window, receiver window}

Question 12

slow start and the TCP sawtooth

Answer 12

to start the flow, we double packets sent each time we receive an acknowledgement - until we hit a loss

Question 13

what are two types of loss in TCP?

Answer 13

1. timeout (serious problem)
   solution: better to start over with a low CWND or else you might trigger a very large burst in traffic
2. triple duplicate ACK (fast retransmit/recovery)
   solution: do a multiplicative decrease and keep going

Question 14

Explain the motivation for Nagle’s algorithm?

Answer 14

Question 15

How does Nagle’s algorithm deal with this issue?

Answer 15

Question 16

In Nagle’s algo, when there is a loss in the network…

Answer 16

more than one small packet is sent per round trip time

Question 17

What is the motivation for delayed ACK?

Answer 17

Question 18

Delayed Ack

Answer 18

Question 19

If B wants to send an acknowledgement to A, but has no data to send to A, what does it do?

Answer 19

Wait to piggyback on data generated within the timeout period, otherwise directly send the acknowledgement back

Question 20

What is the TCP Variant, Fast Retransmits?

Answer 20

Question 21

When does fast retransmit work best?

Answer 21

long data transfers
high window size
low burstiness in packet losses

Question 22

What is slow feedback from drop tail?

Answer 22

Question 23

How does RED address the issues of waiting for timeout or triple ACK?

Answer 23

Question 24

What are properties of RED

Answer 24

Drop packets before queue is full
Drop packet in proportion to each flow’s rate
Drops are spaced out in time
Tolerant of burstiness in the traffic

Question 25

What are the problems with RED

Answer 25

Hard to get the tunable parameters just right
Sometimes RED helps but not always
RED is implemented in practice but often not used due to the challenges of tuning right