William Hay <[email protected]> writes: > On Mon, Dec 07, 2015 at 03:20:08PM +0000, Ondrej Valousek wrote: >> SGE can use CGROUPS for this bookkeeping you mentioned, too. It is >> just a different filesystem, so it is quite simple to figure out >> which process belongs to which cgroup. >> To me, when using cgroups, additional GID should not be needed any >> longer. At least per: >> >> http://blogs.scalablelogic.com/2012/05/grid-engine-cgroups-integration.html >> >> Ondrej > I think the cgroups implementation on SoGE differs from the OGS one. > It doesn't do everything the OGS one does.
I wouldn't know, since it isn't available as far as I know. > Supplementary GIDS are more portable than cgroups and have a natural fit to > the > traditional unix access control model. While cgroups are probably better > than GIDs for killing jobs. [Non-Linux systems have (had?) better ways of dealing with this, of course. cgroups are really not the right answer to most HPC resource management issues as far as I can tell.] > One thing we do with the additional GID is control access to GPUs by chgrping > them from the prolog. While I believe you can do similar tricks with cgroups > its a little more involved. That sort of thing is why I wouldn't eliminate it. cgroups won't help you with filesystem permissions as far as I know, e.g. for access to the job's spool area, which should probably be done that way (but currently isn't). _______________________________________________ SGE-discuss mailing list [email protected] https://arc.liv.ac.uk/mailman/listinfo/sge-discuss
