Messages in this thread | | | From | (Erik Corry) | Subject | Re: Pentium (and Cyrix) DEATH in user-mode | Subject | Page colouring | From | Zlatko Calusic <> | Date | 11 Nov 1997 01:59:29 +0100 |
| |
"David S. Miller" <davem@dm.cobaltmicro.com> writes:
> Time some kernel builds with the following combinations: > > 1) Vanilla-2.1.62 > 2) Vanilla-2.1.62 + page coloring patches > 3) VGER from before I merged in the page coloring stuff > (this is to get a measurement with the "need_resched hack" > on SMP) > 4) Most recent VGER (page coloring + "need_resched hack") > > On non-SMP I was getting 7% (consistantly) faster kernel builds with > the page coloring stuff on sparc64, and this is on boxes where memory > bandwidth is good enough that it shouldn't have shown up that much. >
I also tested Mark's preliminar page colouring patch, but results were different.
I got a small slowdown (< 2%) with the patch applied (638/628 seconds user time, system time didn't change!?).
I was compiling kernel (2.1.62) to test performance. I tried my best to make similar setup for the two compilings.
BTW, I noticed that pages get swapped out rather fast when the colour patch is in the kernel, probably because of that 'weak defragmentation' Mark mentioned.
I experienced the same behaviour with my small patch intended to prevent network blockups I was getting because of memory fragmentation. So I stopped working on that and changed defines in slab.c, so slab allocator doesn't need pages of high order while transferring lots of data via network.
My setup: P133, 32MB RAM (256KB L2, pipeline burst).
-- Posted by Zlatko Calusic E-mail: <Zlatko.Calusic@CARNet.hr> --------------------------------------------------------------------- The Universe is a big place... perhaps the biggest
| |