On Saturday 22 December 2007 01:17, Rik van Riel wrote: > On Fri, 21 Dec 2007 21:52:19 +1100 > > Nick Piggin <[email protected]> wrote: > > BTW. if you have any workloads that are limited by page reclaim, > > especially unmapped file backed pagecache reclaim, then I have some > > stright-line-speedup patches which you might find interesting (I can > > send them if you'd like to test). > > I am definately interested in those. Sorry it took a few days... So I'm testing throughput for the full pagecache lifecycle, insertion to reclaim. Reading from a sparse file much bigger than memory. On my 2 socket, 8 core Opteron setup: Single dd from a single file: vanilla: 687.01MB/s patched: 1037.17MB/s (151%) 8 dds from a single file: vanilla: 1458.04MB/s patched: 5898.65MB/s (405%) Not sure how well that translates to real world workloads, but it might help somewhere. Admittedly some of the patches are pretty complex... Anyway, have a good christmas and new year :) Thanks, Nick
Attachment:
reclaim-speedups.tar.gz
Description: application/tgz
- Follow-Ups:
- Re: [patch 17/20] non-reclaimable mlocked pages
- From: Rik van Riel <[email protected]>
- Re: [patch 17/20] non-reclaimable mlocked pages
- References:
- [patch 00/20] VM pageout scalability improvements
- From: Rik van Riel <[email protected]>
- Re: [patch 17/20] non-reclaimable mlocked pages
- From: Nick Piggin <[email protected]>
- Re: [patch 17/20] non-reclaimable mlocked pages
- From: Rik van Riel <[email protected]>
- [patch 00/20] VM pageout scalability improvements
- Prev by Date: Re: Updated Kernel Hacker's guide to git
- Next by Date: Re: Updated Kernel Hacker's guide to git
- Previous by thread: Re: [patch 17/20] non-reclaimable mlocked pages
- Next by thread: Re: [patch 17/20] non-reclaimable mlocked pages
- Index(es):