HW2 Sample Solution
1/27/06
Grading Criteria
1. Stop and Wait
Example source
Reliable.tar
Example evaluation
Experimental Setup
The network used for this experiment was the CSE wireless network (54Mbps), which is connected to
a 1Gbps backbone. The server used was attu3, and the client ran on my personal laptop. (FC4 Linux,
802.11g wireless, 1.8Ghz Pentium M). I was unable to observe a latency greater than 1ms at any time
during my tests.
Data
By disabling retransmissions, the underlying reliability of the network can be observed. What I found
was that, the network is extremely reliable, and as long as the timeout is greater than the round trip
time, there was no loss at all.
To examine the throughput, a large file (251 kilobytes) was transmitted across the network and two
quantities were measured, the total time to transmit, and the number of retransmissions.
Analysis
Because the network behaved very reliably during my tests, I can only assume that the real unreliability
of the network is being covered up by the lower layers of the network. It should be noted that, on larger
datagram sizes, such as those tested for throughput, the loss increases somewhat.
Throughput was measured in Kbps, by measuring the time the client spent sending and dividing it by
the total amount of bits sent. The throughput increases substantially at high datagram sizes, but some
loss is introduced if the datagram size is too big. This can be seen in the retransmissions graph.
In completing this experiment, I clearly saw how the throughput is severely limited by the protocol of
stop and wait. Since the RTT is so low on the tested network, most of the time is spent waiting around
for acknowledgment packets.
Graphs
2. & 3. Statistical Multiplexing and Book Questions
Solution as PDF or text.
isdal (at) cs.washington.edu
jjohn (at) cs.washington.edu