[Clas_offline] Fwd: [Jlab-scicomp-briefs] batch farm memory requirements

[Harut Avakian]

Dear All,
Please check the actual usage of memory for your farm jobs.
Reducing the requested memory will let more jobs to run (see below 
Sandy's e-mail).
Making the code multi-threaded is another thing we should consider.
Harut


-------- Original Message --------
Subject: [Jlab-scicomp-briefs] batch farm memory requirements
Date: Sat, 1 Nov 2014 09:20:34 -0400 (EDT)
From: Sandy Philpott <philpott at jlab.org>
To: jlab-scicomp-briefs at jlab.org

Dear JLab Data Analysis Cluster Users,

We are looking into ways to adjust fair share on the farm, to include 
not just job slots but also memory footprint -- currently ~400 job slots 
are available, but all pending jobs are requesting too much memory to 
fit in,  because running jobs are using more than 2 GB / core.  Also, 
several users are running not only memory intensive but 2 and 3 day long 
jobs as well.

Please check your actual memory usage upon job completion, and reduce 
your memory request if the memory is not actually being used. If you can 
run multi-threaded code rather than single core jobs, this too will help 
efficiency in overall farm job throughput.  Note that the newest FY14 
farm nodes procured and awaiting arrival have 1.5 GB memory per core.

Let us know if you need any help analyzing your job requirements. And, 
if you have small memory requirement jobs, now's a good time to get them 
in for fast turnaround!

Regards,
Sandy
_______________________________________________
Jlab-scicomp-briefs mailing list
Jlab-scicomp-briefs at jlab.org
https://mailman.jlab.org/mailman/listinfo/jlab-scicomp-briefs