On Wed, Apr 06, 2005 at 07:06:17AM -0400, Alan Cox wrote: > On Tue, Apr 05, 2005 at 11:53:33AM -0700, Jeffrey Buell wrote: > > results are pretty ugly compared to a 2.6.1 install. FC4 ranges from 1.5X to > > 20X slower, while xen0 is about 2X to 6X slower. Is this expected? I'm > > looking for configuration differences now, but I'm wondering if anybody can > > The fork ones are expected yes. The newest kernels have four level page tables > and while this allows for huge 64bit systems it has a cost and there is > optimisation work left to do. They should be getting collapsed away back to 2/3-level PTEs by the compiler in the current builds. I thought that got fixed for 2.6.12-rc1. If it missed it, theres a .12-rc2 update pending, but that won't hit until post -test2 now. > I've not checked the kernel config options but if you look at the src.rpm you > can see what debug is enabled Slab debug is the only thing left on thats really a performance hit. Dave
