Messages in this thread | | | Date | Tue, 11 Dec 2007 15:33:36 +0100 | From | Ingo Molnar <> | Subject | Re: Major regression on hackbench with SLUB (more numbers) |
| |
* Steven Rostedt <rostedt@goodmis.org> wrote:
> Hackbench seems to show this regression the most. In my tests I didn't > see much change with kernel builds and such, but the focus was on > scheduling not memory management. I'll run my kernel tests next for > both SLAB and SLUB and see if there's any difference there.
i just ran various benchmarks on an 8-way (8x 700 MHz Xeon, 4GB RAM):
AVG v2.6.24.slab v2.6.24.slub [ smaller is better ] ----------------------------------------- mmap: 1052.66 1049.33 ( 0%) ctx-2: 4.32 4.30 ( 0%) select: 41.95 43.69 ( 4%) proc-exec: 394.45 391.92 ( 0%) hackbench-10: 1.12 2.99 (166%) hackbench-20: 2.04 6.67 (226%) hackbench-50: 5.03 17.50 (247%)
and hackbench overhead stands out, by a huge margin. Other stuff is within measurement noise. Neither SLUB nor SLAB debugging was turned on, all other debugging options were off too.
Ingo
| |