CS 640, Spring 2014

Project 1 Grading

Grades are posted on Learn@UW

Rubric

Item Points Breakdown
Ping results 2 All results = 2, missing one type of result = 1, otherwise = 0
Ping questions 8 Each question is worth 2 points, except #3 is not worth any points
Traceroute results 1 All results = 1, missing some results = 0.5, otherwise = 0
Traceroute questions 3 Each question is worth 1 point
Point-to-point results 1 All results = 1, missing some results = 0.5, otherwise = 0
Point-to-point questions 4 Each question is worth 2 points
Switched network results 1 All results = 1, missing some results = 0.5, otherwise = 0
Switched network questions 4 Each question is worth 2 points
Network-of-networks results 1 All results = 1, missing some results = 0.5, otherwise = 0
Network-of-networks questions 5 Each question is worth 2 points, except #3 is worth 1 point
Total30

Answers to Questions

Note that other answers that agreed with your results and networking principles were also accepted.

Ping

  1. Yes, the average RTT differs. The differences may be caused by variation in propagation delay (due to different distances between CS and the host) and queueing delay (due to different paths between CS and the host). Transmit delay can also have an impact if the bottleneck bandwidth is different.
  2. Yes, the average RTT differs. The differences are likely caused by transmit delay (it takes longer to transmit larger packets).
  3. The location must be somewhere other than a machine connected to the wired or wireless CS networks.
  4. The average RTT for hosts that are in Wisconsin will be different, because now there may be differences in distance and network paths, which results in different propagation and queueing delay. The average RTT for hosts that are in other states or countries will be about the same between the distance and path will still be largely the same from another machine elsewhere in Madison.
  5. Yes, the first (few) RTTs should be different than later RTTs. The extra delay is due to the time required to perform a DNS lookup. It could also be due to ARP. If you run ping a second time, the first measured RTT should not incur extra delay because the DNS entry and ARP entry are now cached on the local machine.

Traceroute

  1. Yes, the first few hops are the same for hosts outside of campus because the path through the campus network to reach the Internet is the same.
  2. Hosts that are farther away will have more nodes in the traceroute.
  3. The latency for a host should be proportional to the number of nodes that show up in the traceroute, because more nodes in the traceroute (usually) means longer distance (and more propagation delay) and more routers to traverse (and more queueing delay).

Point-to-point

  1. The bandwidth measured by iperf approximately matches the bandwidth of the links. The difference may be caused by: waiting for the medium to be idle so the next packet can be sent, burstiness, or TCP, which limits the maximum amount of data in flight as part of its slow start and congestion control mechanisms.
  2. Yes, the latency is significantly higher when measured while a bandwidth measurement is taking place. This is due the fact that ping packets must be queued alongside bandwidth measurement packets, which increases queueing delay.

Switched network

  1. The latency is the same. In the switched network experiment, iperf is running over a different set of links than ping, so it is equivalent to running a ping measurement over an unused point-to-point link.
    It is also acceptable to say the latency is higher due to additional propagation or queueing delay.
  2. The bandwidth measured by iperf approximately matches the bandwidth of the links. Bandwidth is being measured over separate paths, so each iperf flow can fully utilize the links.

Network-of-networks

  1. The bandwidth in the first experiment is half because the link between S1 and S2 must be shared between the two iperf flows, while the link can be fully utilized by the single flow in the second experiment.
  2. The bandwidth in the first experiment is 50% lower because the link between switches is the bottleneck link in both experiments and this link has 50% less capacity in the first experiment.
  3. The bandwidth in the second experiment is equal because the link between hosts and switches is the bottleneck link in both experiments.