Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
Best practices using clusters
Posted 2010年5月13日 GMT-4 11:45 3 Replies
Please login with a confirmed email address before reporting spam
I would imagine some of you run comsol on clusters already. Say I have several powerful machines laying around the institute. Can I turn them into a grid and run simulations on them? Something like BOINC.
The systems are windows 64bit and some linux. If necessary I can dualboot them into linux after daytime work.
The systems are windows 64bit and some linux. If necessary I can dualboot them into linux after daytime work.
3 Replies Last Post 2010年8月12日 GMT-4 07:25
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
My experience of running such "distributed clusters [ but NOT COMSOL] with a dedicated software my former institution had written "in house" is that it is not worth the effort. The bottleneck I found was the network and associated routing was killing the cluster potential capabilities.
Indeed , to run efficiently an hpc cluster [ regardless of the app] fast network lingk and dedicated switch are needed.
and of course all the nodes need to run the same OS and are controlled by a "head node". it is possible tough [ again not speaking of comsol possible specifics need that will come on top of the general stuff I mention here] BUT it will be a pain in the butt, lot of night and week end time and associated coffee ..and computing performance will be very bad.
the ONLY scenario I can think of where such experience could be useful is that you want to learn how to run comsol on a cluster [ and how to set up and administer a cluster before that...] ....
well in that case have fun it can work ..and you will learn a lot for sure :-).
JF
afterthoughts: it is possible that for running parametric case you could benefit from such a cluster but in principle if you have comsol on all the machines you can write a script [ in python for eg] or maybe with matlab that will distribute the run. the tradeoff is spending time setting up same os on your network and setting up a cluster to benefit from comsol automatic parametric mode vs. distributing the comsol case at system level through a "non comsol" script..
no opinion on what will be best for you in that scenario
Indeed , to run efficiently an hpc cluster [ regardless of the app] fast network lingk and dedicated switch are needed.
and of course all the nodes need to run the same OS and are controlled by a "head node". it is possible tough [ again not speaking of comsol possible specifics need that will come on top of the general stuff I mention here] BUT it will be a pain in the butt, lot of night and week end time and associated coffee ..and computing performance will be very bad.
the ONLY scenario I can think of where such experience could be useful is that you want to learn how to run comsol on a cluster [ and how to set up and administer a cluster before that...] ....
well in that case have fun it can work ..and you will learn a lot for sure :-).
JF
afterthoughts: it is possible that for running parametric case you could benefit from such a cluster but in principle if you have comsol on all the machines you can write a script [ in python for eg] or maybe with matlab that will distribute the run. the tradeoff is spending time setting up same os on your network and setting up a cluster to benefit from comsol automatic parametric mode vs. distributing the comsol case at system level through a "non comsol" script..
no opinion on what will be best for you in that scenario
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Something like BOINC.
I will present a approach on the COMSOL Conference 2010 in Paris. Accepted paper title "ComsolGrid – A framework for performing large-scale parameter studies using Comsol Multiphysics and BOINC".
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Something like BOINC.
I will present a approach on the COMSOL Conference 2010 in Paris. Accepted paper title "ComsolGrid – A framework for performing large-scale parameter studies using Comsol Multiphysics and BOINC".
One more reason to attend the conference :)