Hi, sorry, I can not tell whether it is a common problem (I did not encounter this). But I believed it is important to determine whether it might come from the network issues. F.
2014-12-02 11:16 GMT+01:00, Gupta, Ratnesh (GE Global Research, Consultant) <ratnesh.gu...@ge.com>: > Hi Filip, > > Yes, I tried multiprocessing on single node too. And the problem persists. > Is it a common problem with meep multiprocessing? > > Thanks and Regards, > Ratnesh > > -----Original Message----- > From: Filip Dominec [mailto:filip.domi...@gmail.com] > Sent: Tuesday, December 02, 2014 3:41 PM > To: Gupta, Ratnesh (GE Global Research, Consultant) > Subject: Re: [Meep-discuss] Slow calculation with meep parallelisation > > Hi, I believed you might have problems with the inter-node communication in > the cluster; I wondered whether the problems pertain when the > multiprocessing is run on single node. Maybe you have tried this? > F. D. > > 2014-12-02 10:02 GMT+01:00, Gupta, Ratnesh (GE Global Research, > Consultant) <ratnesh.gu...@ge.com>: >> Hi Filip, >> >> Thanks for the quick reply. I didn't understand what you mean by >> running on single node and using different -np option. Or if you meant >> both separately. >> >> I did try running on single node using '-np 1' with 'meep-mpi'. Also I >> did try with strictly single node using 'meep'. 'meep' and 'meep-mpi with >> -np 1' >> took similar time for a smaller model but both ran out of memory for >> the required geometry size giving the segmentation fault. >> >> I hope I could clarify my problem. >> >> Thanks and regards, >> Ratnesh >> >> -----Original Message----- >> From: Filip Dominec [mailto:filip.domi...@gmail.com] >> Sent: Tuesday, December 02, 2014 2:24 PM >> To: Gupta, Ratnesh (GE Global Research, Consultant) >> Subject: Re: [Meep-discuss] Slow calculation with meep parallelisation >> >> Hi, Ratnesh, >> it indeed seems to be an issue with multiprocessing - did you try to >> run it on a strictly single node? How does it scale on it when you use >> different -np option? >> F. >> >> 2014-12-02 9:18 GMT+01:00, Gupta, Ratnesh (GE Global Research, >> Consultant) <ratnesh.gu...@ge.com>: >>> Hi, >>> >>> I've configured a cluster with meep-mpi. I am able to execute the >>> code. But the code while run with '-np 1' is completely run within >>> 368 seconds where it took 410 seconds while run with '-np 2'. >>> Similarly, the time keeps increasing for increasing number of >>> processors which is not expected. >>> >>> Also, I have a huge geometry which I want to simulate and is getting >>> segmentation fault with single processor. I hoped to run it with >>> multiple processor so that do decrease the probability of >>> segmentation fault. >>> >>> Please help. >>> >>> Thanks and Regards, >>> Ratnesh >>> >> > _______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
