Hi Paul, I notice you have no output file here. I'm not sure how mpich handles buffering of stdout, but that might be hurting you. Try running with '-o /path/to/outputfile'. There might be a temp file to which output is being written. Running 'lsof' on the rank 1 process can tell you this. Output can often be rate-limiting with mpiblast. Do you expect to get a lot of hits? If so, you may want to limit the size of the output file via limiting the number of results returned per query (blastall -b and/or -v) or, more usefull, e-value (-e). What percentage of the CPU is rank 1 using compared to ranks 2-13? -Lucas On Friday, February 03, 2006 at 11:07 -0800, Paul Mc Kenna wrote: > We are trying to get the hang of a small cluster we just built. A > co-worker launched a 10K query against the whole human genome. It has > been running for 3 days now! He had previously launched a 1K query which > something else likely to be going on. Is there anyway to check on > exactly how much of a Blast job has been completed. > > Process currently running on 6 compute nodes, 4 are dual processor > boards. Most have have 2G or more of memory. A quick look of the stats > showed no more than 12% of available memory being used, less on most > machines. Essentially no swap has been used to the best of my knowledge. > > We used the following syntax to launch the job: > > P4_GLOBMEMSIZE=268435456 time /opt/mpich/gnu/bin/mpirun -np 14 > /usr/local/bin/mpiblast -p blastn -d whole_genome.fa -i /home/kieran/rnd.seq > > > Thanks > > > Paul > _______________________________________________ > Bioclusters maillist - Bioclusters at bioinformatics.org > https://bioinformatics.org/mailman/listinfo/bioclusters