Opened 6 years ago

Last modified 2 years ago

#1426 new enhancement

implement RSMAP complex

Reported by: dlove Owned by:
Priority: normal Milestone:
Component: sge Version: 8.1.0
Severity: minor Keywords:
Cc:

Description

http://gridengine.eu/grid-engine-internals/102-univa-grid-engine-810-features-part-2-better-resource-management-with-the-rsmap-complex-2012-05-25
looks a reasonable alternative to SLURM's generalized resources
interface, and possibly more flexible.

Initial work on a skeleton (note TODOs) at
https://github.com/gridengine/gridengine/tree/RSMAP

Change History (3)

comment:1 Changed 5 years ago by michel

This enhancement would be very usefull for us.
Will it be implemented in a near futur ?

comment:2 Changed 2 years ago by dpenner_3vg

Has there been any progress on this? If not, are there any alternatives in Son of Grid Engine that will allow us to specify how many gpu's a job wants/requires and have the scheduler tell the application which gpu's to use? This could be communicated by an environment variable, command line option or similar.

comment:3 Changed 2 years ago by dlove

Has there been any progress on this?

Not as far as I know, sorry.

If not, are there any alternatives in
Son of Grid Engine that will allow us to specify how many gpu's a job
wants/requires and have the scheduler tell the application which gpu's to
use? This could be communicated by an environment variable, command line
option or similar.

You mean if you have more than one GPU and job per host? I vaguely
remember William Hay had something, probably
<https://github.com/UCL/Grid-Engine-Prolog-Scripts> -- I think he
listens. I guess you could do it with a consumable assigned to multiple
queues on the host, since the allocated queue is provided to the job.

[It sounded as though Sun had a scheme for that sort of thing in the
distribution of what became VirtualGL; someone had here, but I had no
luck asking previously, and couldn't find it anywhere else.]

Note: See TracTickets for help on using tickets.