Improve run time on single processor by using mpiexec

Issues and examples using this model (available in the Models Repository)

Improve run time on single processor by using mpiexec

Postby ddarkin on Fri Mar 13, 2015 3:47 am

I'm sorry if this is common knowledge, but I though I'd share this for the benefit anyone who might not be utilising mpiexec commands within windows and wants to improve model run times.

I assume like many users of COULWAVE I am using the code for a university project, and began working with the code with no prior knowledge of MPI based programming. One issue I had with running my models (within windows through running the precompiled .exe file in the command prompt) was that the process never seemed to use more than about 20% of my cpu, and for a 6x6km grid a 1000 second model might take in excess of 35 hours to run.

I've managed increase cpu usage by adding a short line of code before running pcoulwave.exe in the command line:

cd c:\(working folder location)
mpiexec -n 8 -cores 4 pcoulwave.exe
0
0
0
0
0
0
0
0

This has the effect of running 8 instances of the process within windows while utilising all the cores of my i7 processor, and drives cpu usage right up to 100% while the model is running thus turning a 35 hour run into a 7 hour run. Like I said I have no experience of MPI programs and this command will probably vary depending on the processor being used, but its helped me greatly. Bare in mind this changes how the cmd prompt window updates, so it helps to have the model options already chosen by just running the .exe and choosing the model run options prior to running the code above, where you can just type "0, enter, 0, enter..." ect until the program starts running
ddarkin
 
Posts: 1
Joined: Mon Oct 27, 2014 7:50 am

Return to COULWAVE User Discussions

Who is online

Users browsing this forum: No registered users and 1 guest

cron