Messages in this thread | | | Date | Thu, 26 Feb 2009 11:22:36 +0000 | From | Mel Gorman <> | Subject | Re: [RFC PATCH 00/19] Cleanup and optimise the page allocator V2 |
| |
On Thu, Feb 26, 2009 at 01:18:59PM +0200, Pekka Enberg wrote: > On Thu, 2009-02-26 at 11:03 +0000, Mel Gorman wrote: > > On Thu, Feb 26, 2009 at 05:10:27PM +0800, Lin Ming wrote: > > > We tested this v2 patch series with 2.6.29-rc6 on different machines. > > > > > > > Wonderful, thanks. > > > > > 4P qual-core 2P qual-core 2P qual-core HT > > > tigerton stockley Nehalem > > > ------------------------------------------------ > > > tbench +3% +2% 0% > > > > Nice. > > > > > oltp -2% 0% 0% > > > > This is a big disappointment and somewhat confusing that it is so > > severe. For sysbench I was seeing on six different machines; > > > > 50834.14 51763.08 1.79% > > 61852.08 61966.58 0.18% > > 5935.98 5980.06 0.74% > > 29227.78 30167.72 3.12% > > 66702.67 66534.76 -0.25% > > 26643.18 26542.59 -0.38% > > > > So, two smallish regressions but mainly gains. Then again, I'm becoming > > more and more convinced that sysbench doesn't really represent a proper > > OLTP workload. > > > > I'd like to understand more how the page allocator at least was being used > > during your tests. Would it be possible to get a full profile (including > > instruction if possible and the vmlinux file) for both kernels please? > > > > If you can get the profiles, confirm the regression is still there as > > sometimes profiling can alter the outcome. Even if this happens, the > > profile will tell me where time is being spent. > > > > > aim7 0% 0% 0% > > > specjbb2005 +3% 0% 0% > > > hackbench 0% 0% 0% > > > > > > netperf: > > > TCP-S-112k 0% -1% 0% > > > TCP-S-64k 0% -1% +1% > > > TCP-RR-1 0% 0% +1% > > > UDP-U-4k -2% 0% -2% > > > > Pekka, for this test was SLUB or the page allocator handling the 4K > > allocations? > > The page allocator. The pass-through revert is not in 2.6.29-rc6 and I > won't be sending it until 2.6.30 opens up. >
In that case, Lin, could I also get the profiles for UDP-U-4K please so I can see how time is being spent and why it might have gotten worse?
Thanks
> > > > > UDP-U-1k +3% 0% 0% > > > UDP-RR-1 0% 0% 0% > > > UDP-RR-512 -1% 0% +1% > > > > > > Lin Ming > > > > > > > Thanks a million for testing. > > >
-- Mel Gorman Part-time Phd Student Linux Technology Center University of Limerick IBM Dublin Software Lab
| |