> > Any ideas on where to look further? > > I will send you patch for testing. Two more data points: the sock loss is also in test5, but not in 2.2.17pre18 - on the other hand the 2.2 kernel does less than 1000 connections per second, where 2.4-test does 4500... Maybe there is a general problem in mm/slab.c? I remember some change in the SMP case. When did that go in? Patrick - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
