On Thu, Dec 29, 2005 at 05:11:18PM -0700, Hans Fugal wrote: > My dad is having serious come-and-go packet loss issues with his ISP > (a Vernal local wireless setup). I'd like to give him some leverage > with some nice cacti graphs of packet loss, but I'm having a hard > time pinning down precisely what to graph.
I would pay attention to TCP packets sent out wherein the RTT is greater than 1.5 seconds. Just grep the tcpdump output, keeping a hash table of what is sent out vs. what you get back. Keep in mind that TCP has an additive-increase multiplicative-decrease strategy, and, unfortunately, packet loss is the only congestion indicator that TCP offers (which is why such schemes incorporating explicit congestion feedback, like XCP, are a really good idea). So packet loss is very frequent for TCP connections; if you want to be able to differentiate from packet loss due to congestion vs. packet loss due to other reasons, then you are up for a challenge, but it has been done (see papers referenced below). Chances are, your dad's ISP is sending packets through a congested bottleneck, and he has a long latency to get to that bottleneck, and so his TCP traffic is being treated unfairly due to weaknesses in TCP. If you are interested in some interesting reading material on the subject, see: http://www.cs.utexas.edu/users/lam/Vita/IEEE/YangLam00.pdf http://www.acm.org/sigcomm/sigcomm2002/papers/xcp.pdf http://www.acm.org/sigs/sigcomm/sigcomm2004/papers/p581-kim1.pdf Mike .___________________________________________________________________. Michael A. Halcrow Security Software Engineer, IBM Linux Technology Center GnuPG Fingerprint: 419C 5B1E 948A FA73 A54C 20F5 DB40 8531 6DCA 8769 A directive occurred while processing this error.
signature.asc
Description: Digital signature
/* PLUG: http://plug.org, #utah on irc.freenode.net Unsubscribe: http://plug.org/mailman/options/plug Don't fear the penguin. */
