| Hi Jeff,
Would it be possible to generalize the LML driver to work with SGE, but have special code to handle TACC? e.g. it could check for "showq" and use it if found, otherwise default to using "qstat". That way we could use it for both, since a number of people have been asking about SGE support.
Regarding the performance issue. I would suggest changing the hierarchy so that you add another level to avoid drawing all 4000 nodes. If ranger doesn't have any logical hierarchy that makes sense, I'd suggest grouping them in something like 100's.
We've been thinking about how to allow additional drivers to be provided, and this is definitely on the todo list.
Greg
On Nov 29, 2011, at 10:19 AM, Jeffrey Overbey wrote: Hi Greg/Claudia/et al.,
I have a contribution question.
I put together a resource manager and LML driver for a couple of the XSEDE/TeraGrid machines at TACC. I'd like to contribute the LML driver to PTP, but there are a couple of things you might not like about it... :)
1. TACC's machines run Sun Grid Engine, but they have a non-standard "showq" (similar to MOAB) for obtaining the list of jobs. Moreover, they disabled collecting job scheduler information through qstat, which I believe would be the usual way to get this information from SGE. So, the LML driver isn't a generic driver for SGE: it is specific to TACC's installation.
2. The driver does not populate the node display (or whatever you call it -- the boxes that shows up in the System Monitor view) due to scalability issues. For example, ranger.tacc.utexas.edu has 3,936 16-way compute nodes -- a total of 62,976 cores. On my Mac, refreshing that view was locking the UI for about 20 seconds.
So, my question is, (1) should I contribute my TACC-specific SGE LML driver, and (2) if not, have you thought about providing an extension point for third-party plug-ins to contribute LML drivers?
Jeff
_______________________________________________
ptp-dev mailing list
ptp-dev@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-dev
|
