For parallelisation, the training scripts doesn't support anything other than sun grid engine.
The decoder isn't threaded or parallelised. The parallelisation is really simple at the mo - the input is horizontally partitioned before running and farmed out to identically decoding processes. It would be interesting to bring modern tools like mpi/openmp to the project. -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Vineet Kashyap Sent: 05 May 2008 02:16 To: [email protected] Subject: Re: [Moses-support] running moses on a super cluster Sanne Korzec <[EMAIL PROTECTED]> writes: > Hi Sanne/Mailing List This is not a reply but i would like to ask a question to the mailing list related to what Sanne has posted. My University also has a cluster which has a total of 16 nodes (11 x PIIIs, 4 x Celerons and a dual PIII). There is an availability of 30 GB of disk space. The base OS is Linux. Red Hat versions 7.3 are used PBS - Batch system to allow users to submit jobs MPI - The parallel libraries supported by the cluster architecture There is a maximum runtime of 1 week but i think that can be extended. Can i use PBS and mpi to run moses on this cluster? Do i have to do all the 9 steps as outlined in training before i can run moses on the cluster or just use the europarl corpus(config file) to test the parallelization? Also, moses-parallel in perl so will need a compiler for it and i am assuming that all it does is runs the decoded on every procesor? Thanks in advance. Regards Vinee > > > > > > _______________________________________________ > Moses-support mailing list > [EMAIL PROTECTED] > http://mailman.mit.edu/mailman/listinfo/moses-support > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
