At today's meeting with Ketan, we talked about what's necessary to get link characteristics. I think a great way to go about this would be to get ICMP traces instead of TCP or UDP because ICMP does not require both endpoints to cooperate. This way, a single endpoint can gather link statistics without having to worry about setting anything up with the other end. However, this raises the interesting question of whether or not routers will treat ICMP packets differently than UDP packets (which is what RTP is encapsulated in). Ketan suggested I email Tyler Johnson with a status update to find out how well ICMP packets will work, and also what sort of environment we can expect on the enduser systems. Tyler can answer this because many of those clients will be using the NCNI infrastructure to communicate, so the intermediate routers and client environments are known.

On another topic, Ketan also suggested I look into collaborating with a Duke student to run some Chord stress tests in either the DiRT lab or Modelnet at Duke. Using Modelnet makes more sense because it can better emulate a WAN while the DiRT lab is pretty much a heterogeneous, isolated, and dedicated network. I'll look into this at this week's TONIC.

Post a comment