> On Mar 25, 2015, at 9:38 PM, Gilles Gouaillardet > <gilles.gouaillar...@iferc.org> wrote: > > On 2015/03/26 13:00, Ralph Castain wrote: >> Well, I did some digging around, and this PR looks like the right solution. > ok then :-) > > following stuff is not directly related to ompi, but you might want to > comment on that anyway ... >> Second, the running of munge on the IO nodes is not only okay but required >> by Luster. > this is the first time i hear that. > i googled "lustre munge" and could not find any relevant info about that. > is this a future feature of Lustre ? > as far as i am concerned, only Lustre MDS need a "unix" authentication > system > (ldap, nis, /etc/passwd, ...) and munge does not provide this service.
My understanding per the Lustre team is that they can use munge, if configured to do so. Not saying that is the case here, but it is one potential use that would explain the presence of munge on the IO node >> Future systems are increasingly going to run the user’s job script >> (including mpirun) on the IO nodes as this (a) frees up the login node for >> interactive editing, and (b) avoids the jitter introduced by running the job >> script on the same node as application procs, or wasting a compute node to >> just run the job script. > that does make sense not to run the script on a compute node. > but once again i am surprised ... as far as i am concerned, lustre IO > nodes (MDS and/or OSS) do not mount the filesystem > (i mean you cannot access the filesystem as if you were on a lustre client). > of course, you can write your script so it does not require any access > to the lustre filesystem, but that sounds like a lot of pain > for a small benefit. > /* that is specific to Lustre. GPFS for example can access the > filesystem from an IO node */ As far as I know, that is not a functional requirement, but rather a typical installation configuration. In other words, there is nothing preventing you from mounting the filesystem from the IO node - it’s just a common choice not to do so. It’s also a question of how the system designer decided to define an “IO node”. You’ve made an assumption that may not be accurate. > > Cheers, > > Gilles > _______________________________________________ > users mailing list > us...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users > Link to this post: > http://www.open-mpi.org/community/lists/users/2015/03/26539.php
