On Friday 13 February 2004 03:05, TriKster Abacus wrote: [ Snipped ] Have you, who experience the 2.6.x problems, made sure that you haven't made your kernel with framepointers? That can make a difference under constant heavy load (This option is on by default in the kernel debugging menu).
I think the largest difference simply is the fact that 2.6.x kernels use 1 ms runing slices and the vanilla 2.4.x uses 10 ms slices. Task switching is very time consuming, because it invalidates all CPU caches and the more so in the small cached Intel CPUs. Some 2.4.x sources like gentoo set the internal kernel time period from 200 to 500. This means from 5 ms to 2 ms runing slices compared to the 2.6.x kernels 1 ms slices. The 2.6.x kernels slice interval in most cases makes the preemption unnecessary. -- [EMAIL PROTECTED] mailing list
