On Thu, 15 Sep 2011 11:25:29 -0500, Andrew Hastings wrote: > Eric B Munson wrote: >> On Wed, 20 Jul 2011, Andrew Hastings wrote: >> >>> Eric, >>> >>> On 07/11/11 10:51, Eric B Munson wrote: >>>> Transparent huge pages (THP) make using huge pages on x86(_64) >>>> incredibly easy. >>>> However, the standard glibc malloc is not setup to optimixe >>>> allocations to be >>>> fulfilled as THP, relying on khugepaged to scan and merge huge >>>> pages when >>>> possible. This patch set extends the morecore idea that was >>>> originally used to >>>> back the heap with hugetlbfs huge pages to ensure that the heap is >>>> grown in >>>> huge page sized increments. This allows the kernel allocator to >>>> start the new >>>> area as a huge page rather than wait for promotion. >>>> >>>> Eric B Munson (3): >>>> Add support for THP in morecore >>>> Add support to hugeadm for configuring transparent huge pages >>>> Add controls to hugectl for new THP related env variables >>>> >>>> HOWTO | 7 +++- >>>> hugeadm.c | 105 >>>> +++++++++++++++++++++++++++++++++++++++++++++++ >>>> hugectl.c | 17 +++++++- >>>> hugeutils.c | 4 ++ >>>> libhugetlbfs_internal.h | 1 + >>>> man/hugeadm.8 | 35 ++++++++++++++++ >>>> man/hugectl.8 | 11 +++++ >>>> morecore.c | 84 >>>> +++++++++++++++++++++++++++++++++++-- >>>> 8 files changed, 258 insertions(+), 6 deletions(-) >>> Please consider adding some test cases. >>> >>> -Andrew Hastings >>> Cray Inc. >> I agree that test cases are a good thing for new functionality, but >> I don't see >> what could be tested. THP availablility is completely up to the >> kernel, this >> method doesn't guarantee that the heap will be on huge pages, it >> only aligns >> allocations so that if there are THP available, they will be used >> without >> having to wait for khugepaged to merge them. So I suppose we could >> test that >> every call to brk is 2MB aligned, but I don't see how that could >> ever fail. >> Eric > > I agree, it's hard to make the tests self-verifying. > > But I'd find it valuable to have tests that at least exercise the > code paths in the library, to help catch any breakages (eg segfaults) > from future code changes. > > And each test could report PASS if it finds hugepages on the heap, > INCONCLUSIVE if it doesn't; partial information seems better than > none. > > -Andrew Hastings > Cray Inc.
If it's okay with you, I will merge the latest set and send out a patch that just exercises the thp heap code. Thanks, Eric ------------------------------------------------------------------------------ Doing More with Less: The Next Generation Virtual Desktop What are the key obstacles that have prevented many mid-market businesses from deploying virtual desktops? How do next-generation virtual desktops provide companies an easier-to-deploy, easier-to-manage and more affordable virtual desktop model.http://www.accelacomm.com/jaw/sfnl/114/51426474/ _______________________________________________ Libhugetlbfs-devel mailing list Libhugetlbfs-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
