search

jmichel80 / jmichel80.github.io

My GitHub Pages home page
0 stars 0 forks source link

Parallellization of collective variables in PLUMED using openMP #10

Open carajillu opened 9 years ago

carajillu commented 9 years ago

Hi all! I would like to learn how to use openMP and/or other techniques to parallellize the evaluation of expensive collective variables in PLUMED. I would like, for example, to learn how to best use OpenMP for this purpose, and how to select an optimal number of threads for use with a Gromacs simulation

gtribello commented 9 years ago

Hello Thanks for your interest in this meeting and for making a suggestion. Can you possibly provide some more details of what you would like to do? I can tell you already that there are some collective variables in plumed that are already parallellized using OpenMP and I am pretty certain that you can use openMP parallelism with Gromacs+plumed. If I were trying to do this I would run with different numbers of threads and see what number of threads gave me the most nanoseconds per day. I've never really tried though. Also openMP is not really something that I am expert in so it would be useful to know what you want to do before the meeting so I can do a bit of study.

Thanks Gareth

carajillu commented 9 years ago

Hello,

Thank you for your message. I am planning to work on the development of a second version the JEDI collective variable (http://pubs.acs.org/doi/abs/10.1021/ct501072t), for which I will use PLUMED 2. One of the issues of the first version is that it is slow if the CV acts on many atoms, so I was hoping to speed up the code by parallelizing it with openMP.

Thanks, Joan