Hi
I'm looking for a way to distribute the 200,000 CosmoMC samples over 20 slots without using MPI. In fact I want to know is it possible to run 20 separate job each of them with 10,000 (in params.ini) samples, and then collect them and analyse them?
to test it, I did this in a very simple way. I just run 20 independent job, and then collect the results from the txt output files and run them with getdist. The cosmological results were stupid as I expected. It seems that CosmoMC does not generate the samples in the first 10,000 tries over the whole range of the parameter defined range.
Is there a solution to do it without using MPI?
Bests
Sample distribution over slots without MPI
-
Hamzeh Alavi
- Posts: 6
- Joined: October 27 2011
- Affiliation: KIT
-
Antony Lewis
- Posts: 1945
- Joined: September 23 2004
- Affiliation: University of Sussex
- Contact:
Re: Sample distribution over slots without MPI
Yes, but it's not a very good idea usually. What's wrong with using MPI?
-
Hamzeh Alavi
- Posts: 6
- Joined: October 27 2011
- Affiliation: KIT
Re: Sample distribution over slots without MPI
Thanks Antony for answer. The problem with MPI is so simple, our cluster does not support it, and I couldn't convince them to install it on our cluster.Antony Lewis wrote:Yes, but it's not a very good idea usually. What's wrong with using MPI?
Then, how is it possible without MPI?
Bests
Hamzeh