-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi folks,
Slurm, like Torque, sets RLIMIT_DATA to match the mem request of a job. However, on Linux malloc() in glibc will use mmap() for any non-trivial allocation and mmap() obeys RLIMIT_AS and not RLIMIT_DATA. So Slurm's setting has no effect on the child processes. In my Torque installs we (both here at VLSCI and in my previous job at VPAC) patch pbs_mom to set RLIMIT_AS rather than RLIMIT_DATA and that works very nicely on both sites as codes will have malloc() fail and can handle that gracefully rather than get brutally slayed by pbs_mom. Would a similar patch be of interest here too? I know we're going to need it but rather than carry a patch it'd be nice to get this upstream and so I'd like to know what would be required for that - would it be enough to just check we're on Linux and set RLIMIT_AS there (as it matches the intention of RLIMIT_DATA) or would it be better off as a configuration parameter? All the best, Chris - -- Christopher Samuel Senior Systems Administrator VLSCI - Victorian Life Sciences Computation Initiative Email: sam...@unimelb.edu.au 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/ iEYEARECAAYFAlHnmZgACgkQO2KABBYQAh+UFgCdE0DgVYczyeUagzA7vW2Fud4c SMwAn3PI0QfPLtGNz7YItOIOQYHBd2JU =5gTy -----END PGP SIGNATURE-----