Change PBSpro backend to request mem rather than vmem.

[GoogleCodeExporter]

When requesting resources for a job, the PBS pro back end currently appears to 
convert the per-tool "Max memory" parameter into a request for vmem.

On both of the local clusters that we are deploying bioflow against standard 
practice would be to make the resource request against mem rather than vmem. 
For example, a typical resource request statement might look look like this:

#PBS -l select=1:ncpus=8:mem=1GB

Is is possible to either a) set the default to request mem rather than vmem, or 
b) have this as a configurable option?

Thanks,

Cas

Original issue reported on code.google.com by CasSim...@gmail.com on 12 Aug 2011 at 4:13

[GoogleCodeExporter]

Original comment by amacgregor on 16 Aug 2011 at 3:29

• Changed state: Accepted
• Added labels: Priority-High, Milestone-Pending, Component-Backend

[GoogleCodeExporter]

Look into passing arbitrary key value pairs per tool.

Original comment by amacgregor on 16 Aug 2011 at 3:29

[GoogleCodeExporter]

Looking at the code, it appears the ssh+pbspro backend doesn't passthrough 
cpus, mem and walltime at all. This should be fixed at the same time.

Original comment by retrogra...@gmail.com on 19 Aug 2011 at 7:06

[GoogleCodeExporter]

Right, that would explain it. As per ticket 63, I was only guessing what the 
problem was from a qstat of the tasks as I could not see how to view the full 
resource request.  Sorry for the misdirection.  

Original comment by CasSim...@gmail.com on 19 Aug 2011 at 7:20

[GoogleCodeExporter]

This can be changed now using the new submission script template. The mako 
variable ${memory} holds the tools memory setting.

Original comment by retrogra...@gmail.com on 25 Aug 2011 at 4:39

• Changed state: Fixed