Revisiting batch/queue systems: slurm and munge
Hi all,
Last August, I started a short discussion here regarding batch management /
queue / scheduler / resource managment software for cluster computing. I
already mentioned slumrm [1] and munge [2]. The conclusion then was
unsatisfying -- we don't have anything in Debian. At the time, I was unaware
of Gennaro Oliva's mail to debian-mentors [3] and his package snapshots.
To cut a long story short, I built two crude packages today I would like to
offer for co-maintenance. Given my existing 80-some Debian packages, I really
shouldn't take any more on. The crude packages will do for work, so if
nobody has time or energy to pick them up ... I won't push the issue either.
Because the packages deal with resources, authentication, ... they are not
exactly trivial and would need some tender love and care to be done real
well. They mostly autoconf fine, esp munge. Slurm needs a replacement
scripts for /etc/init.d, a few contributed manual pages but nothing major.
That said, I think it would be worth it. I am quite impressed with slurm.
For a quick overview, see the website [1] and e.g. the recent presentation
from 2006 [4] . Slurm is under active development and just released 1.2.1,
it now even has a nice little gtk-based gui. [ Munge is used by slurm and is
a smaller/simpler package. It already detects Debian in its init.d script and
does The Right Thing. ]
Gennaro: Are you still interested in working on this? I could possibly act
as mentor and 'final compiler / uploader'.
Anybody else working on clusters who needs a DFSG-free resource manager /
scheduler?
Dirk
[1] http://www.llnl.gov/linux/slurm/
[2] http://home.gna.org/munge/ -- but really also from llnl.gov
[3] http://lists.debian.org/debian-mentors/2006/05/msg00020.html
[4] http://www.llnl.gov/linux/slurm/slurm_design.pdf
--
Hell, there are no rules here - we're trying to accomplish something.
-- Thomas A. Edison
Reply to: