On Mon, Jul 9, 2012 at 7:17 AM, Eric B Munson <[email protected]> wrote:
>
> I can't help much with checking the actual hardware support, but we are
> currently seeing trouble with malloc backed
> by huge pages. At some point, glibc started ignoring the __morecore hook
> and I have not yet tracked down the how/why
> of it. For testing 1GB pages, I recommend you try one of the functions in
> libhugetlbfs that gives direct access to
> huge pages (see get_huge_pages()). You can check if it is using a 1GB
> page by reading /proc/meminfo while the program
> is active, alternatively there is a helper in our test suite that will
> read and parse /proc/self/smaps to determine if
> a given address is mapped to a huge page.
>
>
I get the same results when I use get_huge_pages() to allocate my memory.
The number of free huge pages in /proc/meminfo goes down while the program
is running, but the number of dTLB misses is nearly as high as when I use
4k pages.
Has anyone successfully used 1 gig pages on this CPU (Intel Xeon E5620)?
Or any CPU?
Lest I sound like a total neg, 2 meg pages are working well, so yay for
that ;-)
------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and
threat landscape has changed and how IT managers can respond. Discussions
will include endpoint security, mobile security and the latest in malware
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Libhugetlbfs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libhugetlbfs-devel
