On Fri, Feb 17, 2006 at 07:22:27PM +0000, Mel Gorman wrote: > On Fri, 17 Feb 2006, Bryce Harrington wrote: > > On Fri, Feb 17, 2006 at 11:03:43AM +0000, Mel Gorman wrote: > > > > By chance do you have a web or ftp site where these patches are posted? > > > > > > I didn't, but I do now. > > > > > > list-based anti-fragmentation > > > o http://www.csn.ul.ie/~mel/projects/patches/brokenout/mbuddy/v22 > > > o Full patch is called full-mbuddy-v22.diff > > > zone-based anti-fragmentation > > > o http://www.csn.ul.ie/~mel/projects/patches/brokenout/zbuddy/v5 > > > o Full patch is called zbuddy-v5-full.diff > > > > Great, thanks! > > > > Thank you for picking them up. > > Great. They will always be in a vN directory where N is the number of > release. One request regarding file names - our tools key off of the patch name to determine which kernel to apply against. Would it be possible for you to create symlinks to the patch that include the kernel or kernels you'd like to test against? I.e., to test zbuddy-v5 against multiple 2.6.16-rc3 kernels, something like: .../zbuddy/v5/ 2.6.16-rc3-zbuddy-v5 -> zbuddy-v5-full.diff 2.6.16-rc3-mm1-zbuddy-v5 -> zbuddy-v5-full.diff 2.6.16-rc3-mm2-zbuddy-v5 -> zbuddy-v5-full.diff 2.6.16-rc3-mm3-zbuddy-v5 -> zbuddy-v5-full.diff Or alternately if you're only interested in results of your patch against a specific kernel, if you include that kernel id in the patch name itself, we can parse it from that. (This is what the nfsv4 guys do. c.f. http://www.citi.umich.edu/projects/nfsv4/linux/kernel-patches/) > > Also, if you know of any regression or performance tests that you'd find > > useful to have run automatically for anti-fragmentation (or other > > memory-related aspects), I could work on setting those up to run on x86, > > and later x86_64 and ia64. > > > > The tests I always run are posted with each patch. On VMRegress, they > correspond to bench-kbuild, bench-aim9, bench-stresshighalloc, > bench-hugetlbcap and more recently bench-hotremovecap. However, I'm > interested in any performance figures or reports you are willing to > generate on as many platforms as possible. Ah, excellent, I probably will have a chance to start looking into this in April (March is looking pretty conference-heavy for me.) We've done automated vmregress runs for things before, although I personally haven't set up modules with it. Looks fairly straightforward though; I'll check back in if I have questions once I have some time to tinker with it. Thanks, Bryce
