- This topic has 5 replies, 2 voices, and was last updated 11 years, 8 months ago by Anonymous.
March 27, 2012 at 5:50 am #1216Anonymous
I’ve compiled Rosetta with MPI option and I am trying to use AbinitioRelax.default.linuxgccrelease by command:
mpiexc -n 8 AbinitioRelax.mpi.linuxgccrelease @flags>log &
and I found that there are eight score for each model(eg: model001 model001 … model001, model002, model002…model002). It seems that Rosetta use eight CPU to generate eight pdb file with the same name and each one overwrite the previous one and we only get the last one in the end…..
I am wondering how to use AbinitioRelax.default.linuxgccrelease correctly so that the model won’t overwrite with each other?
March 27, 2012 at 2:07 pm #6858Anonymous
There are at least a dozen other threads on this, including some of your own: http://www.rosettacommons.org/node/1874
AbinitioRelax is non-obviously not MPI compatible. Nobody is “responsible” for the app anymore, so nobody has fixed it to use JD2 and MPI like most application. It’s perennially on the to-do list.
There is a patch for 3.3 floating around somewhere. I can email it to the account you used to register for these boards if you like. The patch is unsupported and unofficial.
MPI will not make Rosetta faster – so you can just run your 8 jobs in separate folders (and different RNG seeds with -constant_seed -jran #####) to get the same results with less hassle.
March 27, 2012 at 7:38 pm #6859Anonymous
well, but I thought that this “AbinitioRelax.mpi.linuxgccrelease” probably changed the previous case….. since it contains ‘mpi’ in the name…
March 27, 2012 at 7:42 pm #6860Anonymous
That problem I can explain. The mpi internal tag just means that you compiled in MPI – it doesn’t mean the executable uses MPI. For technical reasons, everything must recompile from scratch if you add extras=mpi, and Rosetta runs its usual set of files to compile. Almost all of those files are totally unmodified by the MPI mode, but a few are different, so the whole new compilation is kept separate and mpi is inserted into the filenames to mark which is which.
Sorry if I was short – the lack of MPI support for ab initio is a pet peeve of mine.
March 29, 2012 at 1:12 pm #6866Anonymous
is this problem solved in the latest Version 3.4?
March 29, 2012 at 1:24 pm #6867Anonymous
Nope! That’s part of why it bugs me so much…
- You must be logged in to reply to this topic.