mpool/hugepage: set mntent API instead of manually parsing /proc/mounts

[ggouaillardet]

Refs #1822

[ggouaillardet]

@hjelmn is hugetlbfs upported by Open MPI on OS X ?
in other words, is it ok to disqualify this module if mntent is not available ?

[hppritcha]

I plan to open a PR for the Cray large page support this week.

The manual parsing doesn't work reliably on CLE 6

Am Montag, 4. Juli 2016 schrieb Gilles Gouaillardet :

@hjelmn https://github.com/hjelmn is hugetlbfs upported by Open MPI on
OS X ?
in other words, is it ok to disqualify this module if mntent is not
available ?

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#1846 (comment), or mute
the thread
https://github.com/notifications/unsubscribe/AHdNqiF1o4s63jLhXhnmR5_gC30WFdZgks5qSOiPgaJpZM4JEI5P
.

[hjelmn]

@ggouaillardet OS X has hugepage support using an mmap flag. Problem is that I have not been able to reliably allocate huge pages more than 5 min after a boot. It looks like Apple only allows huge page allocation in a pool that is very quickly exhausted.

[ggouaillardet]

@hjelmn i found that there is no /proc/mounts on osx (at least by default on Yosemite) so #ifdef HAVE_MNTENT_H protecting the subroutine body does not alter the behavior (at least on osx)

if you are fine with that, i will merge this PR

[hppritcha]

@hjelmn i think this needs to be checked on a Cray if in fact you'd like the huge page mpool to work on that system.

[hppritcha]

This fix is needed for Cray KNL systems (or more precisely CLE 6) otherwise we see fairly frequent aborts in MPI_Init (unless one is using ofi mtl). So putting this at milestone 2.0.1.

[hppritcha]

@ggouaillardet I'm going to merge this and test the cherry pick on 2.x. Unfortunately native launch is aprun on KNL nodes is all messed up on master.