Posted by BenjaminMarcusJones on April 7, 2011
Here’s a video demonstrating creation and parallel (local machine) execution of a design space containing 36 simulation runs using self developed code in Python and TRNSYS 17. The module will simulate the batch of 36 using a specified maximum CPU (100% here) and number of parallel threads (4 here). This concept can be generalized to any set ‘A’ containing ‘m’ design variables, each ‘Ai’ having a length ‘ni’, and executed using any “input file” -> “simulate.EXE” style simulation software (“Embarrassingly parallel”).
Code is object oriented, using Python 2.6, with standard libraries.
I also wanted to investigate how my hyperthreaded intel quad core machine would handle parallelization of TRNSYS runs.
We can see that above 4-5 cores (Threads-axis), overall performance remains constant (batch completion time remains 50s, z-axis). Hyperthreading cores therefore have no benefit, even though your TRNEXE thread will show 12.5% CPU time. Finally, the “Batch” module can specify a ceiling on CPU usage (if ie/ you want to use your machine while TRNSYS executes).
Open question: Can TRNSYS run COMPLETELY in the background? I don’t need to see these “calculation” windows all the time…
More info to follow, just a quick post for now!