ORP and Flow Control in Packet Networks

Question 1

What are flow control objectives? 2 examples

Answer 1

Finding a good compromise to maintain fairness (provide required QoS

Examples:
- Throttling connections (minimum rate)
- Keep delay and buffer overflow at a reasonable level i.e imposing a delay on flow to carry as much in consideration of whole network

Question 2

Describe the trade-offs between large and small window sizes?

Answer 2

Small windows: avoid delay/congestion
Large windows: allow full-speed transmission (maximum throughput)

Question 3

What are some drawbacks to using windows in network communications? (2)

Answer 3

• Do not regulate end-to-end delay
• Do not guarantee minimum data rate (e.g voice and video sessions)

Question 4

Describe what node-by-node window flow control is?

Answer 4

• There is a separate window for each virtual cirtcuit (VC) and pair of adjacent nodes.

Question 5

What are draw backs to using node-by-node window flow control?

Answer 5

• In end to end window flow control - window high propagation delay is relative to packet size (=> large window size)
• In node to node window flow control - the window size is small (2/3)

Question 6

What is the goal of a rate control scheme?

Answer 6

• Generate desired rate for various connections

Question 7

What is the idea behind a rate control scheme?

Answer 7

Give each session a guarantee data rate (according to its needs)

Question 8

Describe how the leaky bucket scheme works? Draw diagram

Answer 8

• To join the transmission queue, a packet must get a permit from the permit queue.
• A new permit is generated every 1/r seconds, where r is the desired input rate.
• If there is no more space in the permit queue arriving permit packets are lost
• Arriving packets without a permit are stored in the packets without permits queue - and when matched to a permit, get added to queue of packets with permits.

Question 9

What is the equation for minimising network delay not including a throttling penalty?

Answer 9

D= sum from i,j = n,m of Dij(Fij) = min the sum from i = 1 to L of Fi/(Ci-Fi)

Fi = lambdai/mu
Ci = capacity of link i
lambadi = arrival rate at link i
1/mu = average length of packet

Question 10

What is a trivial response to the network delay minimisation question? Is this valid?

Answer 10

That all flows be set to 0 and that requested flow be 0. No, if we need to carry flow it is obvious that setting it to 0 is non-useful.

Question 11

What is the vague description of the minimisation of network delay when penalty for small rw is added?

Answer 11

Minimise the sum of network delay + imposed penalty delay?

Question 12

When is a penalty delay imposed?

Answer 12

When the flow a link wants to carry is too small

Question 13

What is the formula for the imposed penalty delay function? and draw the graph

Answer 13

ew(rw) = aw/(rw(bar) - yw)) = aw/rw

• aw: a constant
• rw: current that the link wants to carry
• rw(bar) : capacity of link
• yw: traffic we want to throw away

Question 14

What is the equation involving rw(bar), yw and rw? And what do the variables signify?

Answer 14

rw(bar) = yw + rw

rw(bar): capacity of the link (offered traffict)
yw: traffic we want to throw away
rw: traffic that is actually carried

Question 15

What is the sum of all the traffic xi in individual links equal to?

Answer 15

sum from i = 1 to L of xi = rw

rw: traffic carried by network

Question 16

What is optimising flow control all about?

Answer 16

It’s all about controlling how much flow is offered to the network so that the throughput is OPTIMISED!