June 2006 - Cobalt - mailman3.cels.anl.gov

cobalt-0.93pre3
by Narayan Desai 30 Jun '06

30 Jun '06

We have a new prerelease of cobalt up on the ftp site. This version has some bugfixes, and has per-queue submission restrictions modeled in their final format. There is even the start of a resource manager operations manual describing all of the knobs for administrators. This should be close to the final release, which we plan to make next week. -nld

1 1

cobalt-0.96pre2 release
by Narayan Desai 27 Jun '06

27 Jun '06

We just tested out the current state of cobalt-0.96. We have released it as 0.96pre2. This version has a whole lot of changes from the previous version. Theron has been here and busy over the last month, and has made most of these changes. The most notable change is the implementation of real queues. There is now a per-queue data structure that jobs are associated with; this allows per-queue submission filters and policies. Also, users can now receive email notifications upon job start and completion. We have a few more things to finish up prior to release: - because the cqm storage format has changed slightly, we need to write a converter - we need to document the new commands for queue administration - we want to add some more sophisticated admin email notification features - overall documentation updates Basically, at this point, the code is usable, and seems stable through our tests. While it could probably be deployed now, I would recommend waiting for the final release, which we expect to do late this week or early next. Incidentally, I would recommend the use of the trac timeline rss feed to anyone who wants to track cobalt development. You can point an rss viewer at: http://trac.mcs.anl.gov/projects/cobalt/timeline?milestone=on&ticket=on&cha… To see ticket request, source code commits and wiki updates as they occur. -nld

1 0

releasing reservations
by Joy Khoriaty 16 Jun '06

16 Jun '06

Hi, releaseres.py takes the following arguments, releaseres -p <partition> name are these enough to uniquely identify a partition to be removed? can't two partitions have the same name, and also can't the partition to be released be reserved in more than one reservation? Cheers, _______________________________________________________________ Joy Khoriaty tel: (617) 253 8865 MIT fax: (617) 253 8674 Center for Theoretical Physics email: khoriaty(a)lns.mit.edu 77 Massachusetts Ave. web: Building NE25-4080 Cambridge MA 02139 _______________________________________________________________

2 3

ftp download site up?
by Joy Khoriaty 15 Jun '06

15 Jun '06

Hi, Is the the download site up? I can't seem to reach: ftp://ftp.mcs.anl.gov/pub/cobalt/rpms/sles9-ppc64/ On another note, we're running Cobalt 0.95.2-1. Is there a newer stable version? The resrevation dependencies in the current version don't seem to be respected. Cheers, _______________________________________________________________ Joy Khoriaty tel: (617) 253 8865 MIT fax: (617) 253 8674 Center for Theoretical Physics email: khoriaty(a)lns.mit.edu 77 Massachusetts Ave. web: Building NE25-4080 Cambridge MA 02139 _______________________________________________________________

2 3

Number of nodes vs block size on BG/L
by Andrew Pochinsky 15 Jun '06

15 Jun '06

Hi, everyone, I'm wondering, if it is possible to tell cobalt to use not a full BG/L block, but a part of it. I know that one can run mpirun with both -np and -partition specified and with -np less the number of nodes on the block; it works correctly. I tried some time ago giving cqsub a requested number of nodes that is not equal to defined partition size, but it did not schedule such a job. Has the situation changed recently? If not, what would it take to implement this? (Some of our problems would run much nicer if the block size were not a power of 2...) Thanks, --andrew

2 1