Mohamed Awad Posted August 20, 2013 Report Share Posted August 20, 2013 Hi,I have two couples of questions:1. Is GMS/PEST capable to run a model calibration run over several super computers linked across a fast network?, each of these computers with multiple cores. If Yes, is there a documentation for that. Is BeoPest included in GMS or can be integreated with it.2. Is GMS/MODFLOW capable to run a direct run a forward run over multiple core on a single machine or across a netwrok.Thanks for your help. Quote Link to comment Share on other sites More sharing options...
Michael Kennard Posted August 20, 2013 Report Share Posted August 20, 2013 1. Built into GMS is the ability to run parallel PEST on one machine using multiple cores. Beyond that you'd need to set up parallel PEST yourself outside of GMS. BeoPest is not included with GMS and cannot be integrated with GMS. You'd need to save your files from GMS, launch BeoPest, and read the results into GMS.2. GMS ships with a parallel build of MODFLOW, which means the PCGN solver or the SAMG solver (a commercial product), if selected, will run in parallel using multiple cores on one machine.By the way, based on our experience, parallel MODFLOW does not make much difference until you have a large number of cells - over a million. Quote Link to comment Share on other sites More sharing options...
Mohamed Awad Posted August 20, 2013 Author Report Share Posted August 20, 2013 Thank You Michael for your prompot Reply,Regarding point 1, i remmber that Mr. Lemon mentioned to us in a GMS course about a year ago that it is possible to run GMS/PEST across several computers connected over a network by adding the paths to each machine in a certain file.unfortuntely, i am not able to find my notes about this file or maybe i got it wrong then? Quote Link to comment Share on other sites More sharing options...
Alan Lemon Posted August 21, 2013 Report Share Posted August 21, 2013 You can edit the run management file (rmf) and then make sure to start the PSLAVE program on the remote machines before you start parallel pest on the host machine. Quote Link to comment Share on other sites More sharing options...
Mohamed Awad Posted August 22, 2013 Author Report Share Posted August 22, 2013 Hi Alan,Thank you. Quote Link to comment Share on other sites More sharing options...
Mohamed Awad Posted August 22, 2013 Author Report Share Posted August 22, 2013 Dear Alan,Two more questions:Do I have to copy the model on the remote machines at the pre-specified paths in the rmf file?Do I need to install/setup GMS on the remote machines or just I have to start the PSLAVE program?Thank you for your assistance. Quote Link to comment Share on other sites More sharing options...
Alan Lemon Posted August 22, 2013 Report Share Posted August 22, 2013 You will have to copy the MODFLOW input files onto the remote computers.You don't have to set up GMS on the remote computers. When PEST runs it uses a batch file and that batch file will have the path to the MODFLOW executable. When I do this I will usually put the MODFLOW executable in the same directory as my MODFLOW inputs and I will modify the batch file to point to this "local" MODFLOW executable. Then I will copy that folder to the various locations on the network along with PSLAVE. Then I get PSLAVE running on the remote computers. Then I launch parallel PEST on the host and you should be good to go. Quote Link to comment Share on other sites More sharing options...
WetStone (Daniel S Fisher) Posted August 22, 2013 Report Share Posted August 22, 2013 (edited) Does MT3DMS use the parallel computation on multiple cores? If not, do you have plans to add that capability? Those transport run times are substantially longer than any flow modeling. Edited August 22, 2013 by WetStone (Daniel S Fisher) Quote Link to comment Share on other sites More sharing options...
Alan Lemon Posted August 22, 2013 Report Share Posted August 22, 2013 I have mentioned this to the MT3D author multiple times but I have not heard or seen any developments along that front.I have seen work done on make the reaction component to transport parallel but usually the advection and dispersion are the time consuming processes. Quote Link to comment Share on other sites More sharing options...
WetStone (Daniel S Fisher) Posted August 23, 2013 Report Share Posted August 23, 2013 That's kind of surprising considering the work done 13 years ago: Parallel Processing of a Groundwater Contaminant Code by R. C. Arnett and L. E. Greenwade, May 22-26, 2000 at the 42nd Annual Cray User Group Meeting. They did determine that ~95% of the cpu time was spent in the advection calculation. Maybe they have completed the code optimization themselves. Get the pdf article here: http://www.google.com/url?sa=t&rct=j&q=&esrc=s&frm=1&source=web&cd=4&cad=rja&ved=0CEEQFjAD&url=http%3A%2F%2Fwww.inl.gov%2Ftechnicalpublications%2FDocuments%2F2690197.pdf&ei=5b8XUtSiEbih4AO2qIDIBw&usg=AFQjCNHVz9uVzfYz1e8yoF5tuana0N1Pnw&sig2=nf6u8CayMJBCiYWwhkUn1w Quote Link to comment Share on other sites More sharing options...
WetStone (Daniel S Fisher) Posted August 23, 2013 Report Share Posted August 23, 2013 Also, Visual Modflow claims to have parallel capability with MT3DMS. Is this merely with the reaction calculation? Quote Link to comment Share on other sites More sharing options...
Alan Lemon Posted August 23, 2013 Report Share Posted August 23, 2013 DanI wasn't aware of either of the items that you posted. Maybe I am mistaken in my understanding of MT3D parallel capabilities. I will check the the MT3D author and see what I can find out. Quote Link to comment Share on other sites More sharing options...
WetStone (Daniel S Fisher) Posted August 24, 2013 Report Share Posted August 24, 2013 Great. I'll be watching to see any new developments! Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.