Archive for October 1st, 2004

Late this afternoon, after a good meeting with Steve Kleiman and Kimberly Melvin of NetApp, I went with Fred, Josh and TJ to the CS department to hear my SILS collegue Dr. Cathy Blake give a talk about her research in datamining in text databases (articles on breast cancer and on impotence for her testing but hopefully, and it seems from her presentation, generalizable to other subjects). Her Information Synthesis approach was the basis of her dissertation. Thanks to her clear presentation, I think even I get it.
One unexpected and pleasant outcome of going to listen to Cathy talk was a discussion with Don Smith and Jasleen Kaur stopped us to chat about their network analysis work. They have been looking at TCP traffic, timeouts and drops on traffic outgoing from the UNC campus network. As you may know, dear reader, ibiblio.org accounts for about half of that traffic. But our per cent dropped traffic while only about 1% is 10 times that of the rest of campus (theirs being a paltry 0.1% except for residence halls which are capped to control the pressure that traffic from those machines would put on the campus backbone.) Why do we drop so much more? Because we too cap our traffic, reluctantly but of necessity, especially on the machines that distribute large high demand files that are available elsewhere on the Net such as Linux distros and our SourceForge mirror as well as the music, for which we encourage the use of BitTorrent (actually BitTorrent would be great for all high demand large files if we can train folks to use it — OSprey?).
But what is most interesting is that the above are only guesses; we have no assurance (or insurance) that our kernel level traffic shaping is doing exactly what we want it to do. We have some pretty good evidence based on performance and the shapes of the traffic charts, but Jasleen and Don could tell us a lot more and in doing so would aid their own research. At least that’s what we all decided. I think we have the basis of some interesting research partnership tied to the public service mission of ibiblio.org. Update: John Reuning sent me the how-to link for the ibiblio QoS set up.
Don and Jasleen’s next research plan has to do with long haul connections grabbing large files. We’re up to be measured there too since folks in Uzbekistan need their Gentoo!
A separate pleasure was a chance to agree with Ketan Mayer-Patel that watching Chan is missing is one of the best ways of understanding the complexities of the Chinese-American cultures in the Bay Area. And it’s a wonderful film to watch too.

Comments No Comments »