Ryan Wilson
CS 244
Assignment 1

Instructions:
To reproduce the graphs and their corresponding results, run 'sudo ./run.sh'
The results for a router buffer of 20 packets are in bb-q20/
The results for a router buffer of 100 packets are in bb-q100/

The time measured for h2 to fetch the /http/index.html file from h1 via curl is
as follows.
Note that the times were measured three times during the 200 second test,
approximately at 50 s, 100 s and 150 s:

With a router buffer of 20 packets:

Average fetch time: 2.506333
Standard deviation of fetch times: 0.054738

With a router buffer of 100 packets:

Average fetch time: 8.394333
Standard deviation of fetch times: 0.338964

Questions
1. The packets when the buffer is 100 packets long will wait in the queue 
instead of being dropped. Thus, TCP will continue sending packets with a larger
window, further bloating the buffer instead of reducing its window.

When the buffer is 20 packets, packets will be dropped and TCP will reduce its
window. Thus, one (or a few) packets must be resent, which is less overhead
than a large number of packets waiting in the buffer.

2. The EC2 instance has a maximum transmit queue length of 1000 packets with a
MTU of 1500 bytes.

The maximum time the packet could wait in the queue is
(1000 packets * 1500 bytes / packet * 8 bits / byte) / (100 Mb / s) = 12 ms

3. RTT = RTT_min + current_buffer_size / bottleneck_link_rate
In this situation, RTT = 20 ms + current_buffer_size / (1.5 Mb / s)

4.
- Reducing the TCP timeout value would prevent bufferbloat as packets stuck in
long router queues would be considered dropped and the TCP cwnd value would be
reduced.
- Setting custom buffer length values in routers and switches based on the
speed of the local network. Many IT administrators leave the router buffers as
the default value.