On 2008-12-16, Stephan A. Rickauer <stephan.ricka...@ini.phys.ethz.ch> wrote: > On Mon, 2008-12-15 at 13:04 +0000, Stuart Henderson wrote: >> On 2008-12-15, Toni Mueller <openbsd-m...@oeko.net> wrote: >> > On Mon, 15.12.2008 at 10:14:41 +0200, Jussi Peltola <pe...@pelzi.net> >> > wrote: >> >> IME forwarded packets seem to somehow have a higher priority than >> >> self-originated traffic in most OS's; don't know why this is, just a gut >> >> feeling. >> > >> > I guess that this is true. In any case, if he would be able to maintain >> > a bandwidth difference between the routers and his uplink, things >> > should start working again. >> > >> > The bandwidth difference could probably be achieved by trunking. >> >> Depending on how the hash function works out, trunking might not >> help there. >> >> What isn't clear yet is whether the problem is caused by the *link* >> being overloaded, or the *firewalls* being overloaded. Stephan, it >> might be interesting to run systat vm .2 on an active firewall >> while the big TSM transfer is taking place, look at cpu use, >> interrupts/sec etc. > > As soon as it happens again I will report all the details. > (Un)fortunately, it doesn't occur too often and I can't simulate that > very easily. > > However, I do remember that interupts were >12000/s, which was mainly > due to em0 and em2 forwarding the traffic (~6000/s each). The cpu load > was ~70% - 80% (it's a Pentium 4, 2.66GHz). The bandwidth utilized was > around 280MBit/s. This all leads to my assumption that rather the > machine was overloaded and not the link. >
You want -current on it then.