Re: job scheduling
Am Sonntag 07 September 2008 20:33:03 schrieb Michael Loftis:
> --On September 7, 2008 12:02:49 PM +0200 Stefan
> <firstname.lastname@example.org> wrote:
> > Am Samstag 06 September 2008 22:39:13 schrieb Michael Loftis:
> >> If youlet us know what your'e trying to do maybe we can help better.
> >> There are quite a few 'types' of job scheduling structures. some, like
> >> condor, are designed to try to distribute jobs, some are cluster
> >> management systems designed to make sure all machines of class X run
> >> script Y. etc.
> > Yes, Im new to this szenarios. Yes what I need at first are job
> > scheduling structures and cluster manangement are also interessting.
> > After abit searching I found
> > http://jobscheduler.sourceforge.net for job scheduling. That looks very
> > good, open source and it supports mysql.
> > Im thinking also about a cluster with multiple machines where I can run
> > distributed jobs, so Im scaleable for greater processing mechanism.
> > btw. Is it better to run batch- jobs on a parallel- Cluster like
> > beowolf?
> Very application specific. There's no one answer. For some applications
> it's better to split the workload up into units, distributing units to each
> compute node (rendering computer graphics and animation, certain types of
> number crunching, lend themselves well to this) -- that's for workloads
> that a single job can take a single node days, weeks, or months. This
> requires an application that understands how to be split up like that. In
> workloads where the compute effort on each job is low, but there are many
> jobs, then scheduling a job as a whole makes more sense, spreading otu the
> many jobs across the available compute nodes. Job/Batch scheduling is like
> a non-interactive (or...less interactive) version of load balancing, though
> load balancing is part of job scheduling. Load balancing by itself can
> refer to other things (like HTTP or SMTP transactions being spread out over
> a cluster of identical machines).
okay, thanks. You mean sometimes its better to use load- balancing rather than
parallel- clustering? Or use it in combination? How to made this disition?
Are there anywhere some parameters I can use?
I find condor also a good solution.
Some of you use "condor" or "jobscheduler" in production and can tell me about
his/ her experience ?