Check to ensure that OMPI was built with the required support - if you run ompi_info, do you see an output line that indicates the RAS tm module was built?
If not, then that is the problem. > On Dec 3, 2015, at 10:36 AM, Gregory Cary <gregc...@andrew.cmu.edu> wrote: > > Hello, > > I'm having difficulty running mpiblast using the code pasted below. The goal > is to run the blast search in parallel on 3 different nodes (2 processors, 8 > CPU cores, and 24 GB RAM per node). Whenever I run mpiexec without the -n > parameter I get the error message: "Sorry, mpiBLAST must be run on 3 or more > nodes". Whenever I run mpiexec and pass it "-n 24" the job runs, but it takes > and excessive amount of time (>8h to search 10 records) and only one of the > three designated nodes is active (memory usage, cpu %, network I/O) during > the job while the other allocated nodes are quiet. > > What am I doing wrong here? It seems as though mpiexec isn't getting the > relevant info about number of allocated nodes and/or the identities of the > specific hosts from the PBS scheduler. Can anyone help me out with this? > > Much appreciated! > ---------------------------- > Greg Cary, Ph.D. > Carnegie Mellon University > Dept. of Biological Sciences > Hinman Lab > 412-268-9290 > > #!/bin/sh > #PBS -N MPI_BLAST -e error_log.txt > #PBS -l nodes=3:ppn=8 -n > #PBS -l walltime=24:00:00 > #PBS -M gregc...@andrew.cmu.edu -m abe > #PBS -q pool1 > > mpiexec\ > mpiblast -p blastp -d nr -e 1e-5 -m 7 -v 10000\ > --removedb\ > --time-profile=time-profile_1.txt\ > -i ~/fasta_files/test2.fa\ > -o ~/fasta_files/blast_results/mpi_test1_10recs.xml > _______________________________________________ > users mailing list > us...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users > Link to this post: > http://www.open-mpi.org/community/lists/users/2015/12/28128.php
