[Moabusers] Control/Monitor node power states

[Chris Samuel]

/* Catching up on email, yes, I know, this is very old! */

On Friday 31 March 2006 08:02, Caird, Andrew J wrote:

> I'm not that familiar with IPMI, can it do clean shutdowns, or is it the
> equivilant of hitting the power button?

IIRC it's effectively a remote power switch, so you'd need to get the node to 
do a proper shutdown first.   I know with Fedora the shutdown 
variant "poweroff" seems to work pretty well to power off nodes (including 
SMP), something that RHEL can't do, so you'll need at least the ability to 
augment that with IPMI/xCat/CSM..

> It still seems to me that the resource manager is somehow closer to the
> hardware than the scheduler, and doing this in Torque makes more sense.
> But I'm no software architect, so there you have it.

I guess the advantage that Moab has is that it knows when queued jobs are 
likely to get run, so it might know there's a large job that can't get on for 
2 days, so shutting down those spare nodes for 24 hours won't hurt at the 
moment.

The issue for me would be the fact that you probably don't want to start 
pbs_mom automatically on boot, because there's precious little to distinguish 
a clean reboot for this from a node that is spuriously rebooting because of a 
dodgy power supply, taking down jobs with it each time it does..

cheers!
Chris
-- 
 Christopher Samuel - (03)9925 4751 - VPAC Deputy Systems Manager
 Victorian Partnership for Advanced Computing http://www.vpac.org/
 Bldg 91, 110 Victoria Street, Carlton South, VIC 3053, Australia

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/moabusers/attachments/20061112/ac1f3e76/attachment.bin