Veda2.0 Released!


Long model generation time
#1
Hi, 

I am struggeling with long generation time for RUN file and LP model. My model is very simple at the moment, with 3 regions, a few technologies and some trade processes. 
As seen from the screenshot attached, it takes more than 14 minutes just for the model generation to finish beofre starting to solve the problem. 

Any thoughs on why this is happening and how to speed up the time?


Attached Files Thumbnail(s)
   
Reply
#2
Thanks for the very intriguing problem report.
I would be grateful if you can provide the TIMES input files for this case (*.DD and *.RUN) plus the listing file (*.LST), to investigate the problem. If that would be possible, just pack these files (e.g. zipped) and you could send them to me via Dropbox, Google Drive or equivalent, by giving me the download link (in a private message if you like).

Thank you.
Reply
#3
Thanks very much for the model input files!
Without them I would not have been able to give any help, but now I was easily able to find the culprit:  The levelization of IRE_FLO had an old performance bug (that part of the code was 13 years old). The good news is that this bug only affected performance.

After fixing that performance bug, the model preprocessing time (before generation) was reduced from 310 to 14 seconds. However, the model generation time is still notable, 160 seconds, but this is not too bad given the size of the model (3 million equations, 12 million non-zeros). The generation "speed" is thus 1.2 million equations per minute, which I think is reasonable. But there is an ongoing ETSAP project looking at the performance, and may achieve some improvements. Anyway, the total GAMS time before the solve was now 14+160=174 seconds.  The fix will be in the next release.

I also see that you are using an oldish version of TIMES (v4.5.6), which also gives you somewhat poorer performance, and so suggest to upgrade.
Reply
#4
Smile
Reply
#5
Dear Antti,
Thank you for the help of finding the performance bug!
Looking forward to seeing the results of the ongoing ETSAP project.
Will the fix be in the upcoming 3.0.6.0 version that will be released next week?
Reply
#6
Yes, we will try to achieve some additional improvement within the ETSAP project. Can I share the input files for your model case with the project partners (GAMS, PSI)?  The current fix will be in the next TIMES release.
Reply
#7
You are welcome to share the input files with the project partners.
Regarding the new TIMES release, do you know when this is scheduled? In the meentime, is it a way that I can get around this performance bug by deactivating some of the input, so that the preprocessing time reduces?
Reply
#8
The next release would be within some weeks' time. In the meantime, you could remove your IRE_FLO parameters from the model (if necessary, define the losses with IRE_FLOSUM instead). And upgrade to the latest version, v4.7.8.
Reply
#9
Will do, thanks for the help!
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)