Matt Ismail <[log in to unmask]> writes:

> May be worth having a quick discussion on scope for this focus group.

Yes.  I'd been intending to ask about the focus of the group, and it's
difficult to decide whether to attend a meeting without knowing what
it's about, though it's rather late in the day.  Most of what's been
said so far hasn't really been about scheduling, but more general
resource management, like checkpointing.

Actually, could we also clarify who's actually meant to be involved?
Should it include someone like me who doesn't normally attend SIG
meetings or know much about it, but has to do the configuration?
(Excuse me if not!)

> I'm thinking specifically that some schedulers can hook
> into various provisioning tools to support things like VMs, green
> provisioning or cloud, which might mean we also need to be concerned
> with cluster systems management. 

I'm puzzled why people separate scheduling/schedulers from management at
all, especially as it isn't a separate component in SGE at the user
level, for instance.  Isn't it more useful to talk about Distributed
Resource Manag{ers,ment} to cover most aspects of actually running jobs,
which seems to be what people are interested in?  E.g. SGE is trivially
converted to the basis of the `XGE' system for scheduling Xen instances,
and the Service Domain Manager components for those buzz words is quite
closely tied to SGE, even if, say, OpenNebula isn't.  (Is the `C' word
on-topic for hpc-sig?)

There's definitely a place for sharing experience of scheduling policies
specifically, though, for those of us without much experience.  It's
probably more generally applicable than discussing provisioning tools
which may not be relevant to people with stateless clusters.

It would be useful to do a comparison of DRMs, especially in the light
of the complaints about SGE, unless anyone knows of a good one.  If
nothing else, it might clear up some confusion like that around SGE's
queues being execution contexts rather than waiting contexts.