cribes how to setup the linking with openmpi, on a p6 575
system, SLES 10 sp2, and libhugetlbfs.
http://tinyurl.com/ch82mn
Bill Buros
--
Create and Deploy Rich Internet Apps outside the browser with Adobe(R)AIR(TM
Mel Gorman wrote:
> HUGETLB_MORECORE currently exists to allow glibc to back malloc() with
> large pages instead of small pages. However, not all applications use glibc
> malloc() nor is it always desirable to back malloc() with huge pages. There
> exists a requirement that a hugepage-aware applica
An easy example is John McCalpin's stream program. Works across
platforms.
http://www.cs.virginia.edu/stream/
We did a quick wiki page example for Power systems at
http://www-941.ibm.com/collaboration/wiki/display/LinuxP/Tuning+stream+with+libhugetlbfs
The wiki page highlights the fact that
> eh?As a user, would does this do?It lets me skip the linking
step?
Sort of. So, if you're a user and you have a binary with a large BSS,
and maybe don't know if you want to try and relink the binary, but would
like to see what is possible, performance-wise, if you were to relink
the bi
Add support for partial segment remapping, where non-relinked binaries
can try to take advantage of libhugetlbfs' segment remapping code. By
LD_PRELOAD'ing the library and specifying the __executable_start address
in the HUGETLB_FORCE_REMAP variable, the library mimics the behavior it
would have
Slowly catching up... the libhugetlbfs version 1.0.1 update was tested
against the SPEC.org published results for SPECcpu2000 (the publish used
version 1) and as expected, we saw no performance regressions.
For the results, see http://www.spec.org/cpu2000/results/res2006q4/ and
search for SLE
I might add... The hugetlbfs mount point should only be accessible to the specific
user or group which needs to use hugepages. That is, the hugetblfsdirectory / mountpoint should be owned by the user and group inquestion and have permissions 0600 (rwx--) or, if multiple usersin the group need
Steve Fox wrote:
>docs: Add information on HUGETLB_DEBUG
>
>
just an observation...on SLES 10, one very common application
segfault is due to ulimit -s being defaulted to 8192. Not sure if
the debug messages spit this out, but we always tell our users to set
the stack to unlimited
On 10/10/06, David Gibson <[EMAIL PROTECTED]> wrote:
Incidentally have people been running the testsuite routinely? For meon current mainline it now produces many errors, and crashes themachine (POWER5 LPAR).Our
performance testing is focused primarily on the distro kernels (SLES
10, OpenSuSE, RHE
> You should be able to 'yum install libhugetlbfs' on Fedora Core 5 and> get the package from the Extras repository (which is enabled by
> default).Hadn't used the Extras or yum before. Pretty niceAnd interesting. It says the kernel with Fedora Core 5 (2.6.15-1.2054_FC5) is in conflict wi
On 8/4/06, Nishanth Aravamudan <[EMAIL PROTECTED]> wrote:
On 04.08.2006 [17:07:42 -0500], Bill Buros wrote:> I installed fedora 5 on an x86-64bit system, got preview 3, installed> that and am trying to link stream.c with gcc. But it's not finding> /usr/local/bin/ld.hugetlbfs
I installed fedora 5 on an x86-64bit system, got preview 3, installed
that and am trying to link stream.c with gcc. But it's not finding
/usr/local/bin/ld.hugetlbfs
I've tried
# gcc -Wl,--hugetlbfs-link=B stream.c -o stream-lp
/usr/bin/ld: unrecognized option '--hugetlbfs-link=B'
Then
# gcc -B
12 matches
Mail list logo
