Is there an easy way via python scripting to use multiple cores on a single computer for PSSE Dynamics Run. It is taking too long to complete dynamics runs, and I suspect the default system is only using a single core. Any help will be appreciated.
Python also has the ability to split work across multiple cores. Some sample code below on how we run multiple dynamic simulations on the same machine in parallel. Be careful on how many workers you add. The problem can become I/O with all the data from the multiple runs and running 4 in parallel will not be 400% faster than running 1. You get a diminishing rate of return the more cores you add. With this code, you can easily play around with how many cores you want to use and find the optimal number for your machine.
Sample code:
import multiprocessing
from multiprocessing import Process
from multiprocessing import Pool
workers = multiprocessing.cpu_count() - 1 # Run on N-1 cores on local machine
p = Pool(workers)
p.map(run_test, tests) # run_test is a function you call in parallel, tests is an array of data that gets passed to the function (could be case name, snapshot, fault name, etc...)
def run_test(test):
# Run you stability run here...
Check the post "Python Parallel Dynamics Simulations", with a link to my python tool MPjobs.
Usually the OS control what task goes to what CPU. In PSSe dynamic runs of small networks, the communication needs imposed by the OS usually are bigger or comparable with the actual running time. Parallel time gains are expected for large system or longer run times. BTW, your scenario seems a 1D.
1D studies allows 1 independent variable (Xvar) in a study, like multiple contingencies, but everything else the same or multiple cases with same contingency. For multiple cases, multiple contingencies, use the 2D.ini example, where 2D studies allows for 2 independent variables (Xvar, Yvar).
Thanks jconto! I'm now using multiple cases and contingencies using mp2D.ini. I'm running dynamic simulations for large networks and still noticing that the runs are happening one by one as opposed to running in parallel. Do you think I'm missing something? Your help is much appreciated!
It seems your runs are in "mptype=MPP" mode, the default parallel mode, an entry in the INI file. The second option is "mptype=MPAA". In MPAA, a run is given to available CPU until all jobs are done. In MPP, jobs groups are formed before each group is given to a CPU, runs are 1x1 within a group.
Thanks jconto! This solved the problem of running multiple instances simultaneously. Appreciate your help on this. I'm now running into issues where after a few runs (2-3 simulations later) it gets stuck somewhere during initialization and only works if I restart mpjobs again. Did you run into this?
You may know this already, but Siemens sells a "Parallel Dynamics Module," which I'm sure will fit your needs. There is a 1-month free trial available by the looks of it. I'm sure it's quite expensive after that.
I couldn't tell you if there's a good way of parallelizing a dynamics run with Python.
Asked: 2019-08-26 13:06:01 -0500
Seen: 1,210 times
Last updated: Aug 29 '19
whit loves you. Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
Thank you all for your responses. This is adequate information for me to get started.