1

Closed

Add option to specify the number of threads

description

When running in a shared or clustered environment, the CPU can become over-subscribed as FastLmm presumes it can use all the CPU resources.  (MKL version)
Is there a possibility to specify the number of threads used by FaSTLMM?
Or to restrict the process to one core?
Closed Apr 17, 2012 at 11:53 PM by bobd00

comments

bobd00 wrote Mar 27, 2012 at 6:00 PM

Can you give me some idea of the issue you are trying to address so I can prioritize this?
I think we can do something, but under the covers the math library (MKL or ACML) exploits the cores available. w/o looking a docs, I believe we can influence the core usage.

costin_leu wrote Mar 28, 2012 at 1:18 PM

we are using fastlmm in a linux cluster enviroment. the number of CPUs used by the software (fastlmm) must be limited to a specific number (for example four or eight) to get not in conflict with other programs on the same cluster node. In one specific case two parallel fastlmm runs required together 17 CPUs on a 16 CPU node leading to a CPU overload. Furthermore because different calculations are running parallel on nodes its neccesarry to restrict the maximum node number used per fastlmm run to preserve a stable environment. Idealy there would be a input option where the user can enter the maximum number of CPUs used, to syncronize this with the number of CPUs reserved for the calculation.

wrote Apr 2, 2012 at 6:23 AM

wrote Apr 10, 2012 at 5:36 PM

-MaxThread <int> command line option checked in and released as part of v1.09
Currently this is only effective with the MKL library version and ignored with the ACML version.  The MaxThreads option is passed directly to MKL.

wrote Apr 17, 2012 at 11:53 PM

wrote Feb 13, 2013 at 11:40 PM

wrote May 16, 2013 at 5:24 AM