Homework 3

Due Oct 28, 2008 (by 5pm)

Note that the solutions will be posted immediately after the 5pm deadline and thus, no late submissions will be accepted.

  1. AIMD: We discussed the additive increase and multiplicative decrease (AIMD) technique for congestion control in class. Using the same graphical technique demonstrate whether or not other alternatives such as MIMD, AIAD and MIAD always converge to a fair state. Also, consider another alternative where increase is additive, however for decrease the connection throughputs are reset to a small constant value -- same for all connections (similar to TCP slow start with window size = 1). Does this technique always converge to a fair state? Explain.

  2. Solve Problem 27 (3rd Ed) or Problem P33 (4th Ed),  Chapter 3 from textbook.  ["Consider the following plot of TCP window size .."]

  3. Solve Problem 31 (3rd Ed) or Problem P38 (4th Ed), Chapter 3 from textbook.. [This is a two part analytical question to determine loss rate L.]

  4. Solve Problem 32 (3rd Ed) or Problem P39 (4th Ed), Chapter 3 from textbook. ["In our discussion of TCP futures in Section 3.7 .. "]

  5. Solve Problem 33 (3rd Ed) or Problem P40 (4th Ed), Chapter 3 from textbook. ["In our discussion of TCP congestion control in Section 3.7 ..."]

  6. Study the Fast Recovery algorithm in TCP Reno carefully, particularly how the congestion window is adjusted. Assume that one single packet is lost and dupacks will arrive at the sender as expected. Assume that right before the 3rd dupack, the window size is W. What will be the window size at the 4th and 5 dupack?  Suppose, now the ack due to the retransmitted packet comes back. What will be the window size?

    Can new data be sent at all before the retransmitted packet is acked? Explain.

  7. As a follow up to the previous question, it turns out that TCP Reno works well when there is a single packet loss inside a window. But with multiple losses within the same window, it is not quite as effective. Argue why? A modification TCP NewReno has been proposed to counter this and is implemented in many modern TCP stacks. Study the NewReno modification. Argue why it could be effective with multiple losses within the same window.