-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi folks,

Our Xeon Phi nodes are all diskless and sadly Intel's libraries hard
code the path /tmp/coi_procs as a place to store files to push to the
Xeon Phi cards.  This means after a few processes /tmp gets full of
copies of MKL, etc. :-(

We had written a hack into our Slurm prolog so that it would create
/tmp/coi_procs/$CARD (where $CARD is 0 or 1 or both) as symlinks to
the users scratch directory for the job, but it turns out that won't
work as the only way to find out which Phis have been allocated is via
$OFFLOAD_DEVICES and that does not appear to be set during the prolog.

Any ideas how the prolog can determine which Phi card(s) have been
allocated to the job?   "scontrol show job" will only show whether 1
or more have been requested.

Thanks!
Chris
- -- 
 Christopher Samuel        Senior Systems Administrator
 VLSCI - Victorian Life Sciences Computation Initiative
 Email: [email protected] Phone: +61 (0)3 903 55545
 http://www.vlsci.org.au/      http://twitter.com/vlsci

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/

iEYEARECAAYFAlIVrOcACgkQO2KABBYQAh/jIACbB133fi1nVfTlhoKM9jUFofTq
nvcAn1LMPXqjDU5Oc66OJbVHyVu/H1Re
=Ekhd
-----END PGP SIGNATURE-----

Reply via email to