[Bioclusters] MrBayes issues in parallel

[Srigurunath (Ecap) Chakravarthi]

On Tue, 11 Jan 2005, Matthew King wrote:
> 1. when I use the nolocal argument with mpirun MrBayes will run through
> the analyses twice. If I do not use the nolocal argument it runs
> through only once. i use the command
> :mpirun -np 4 -nolocal -machinefile
> <pathtomachinefile>/machinefile.freebsd.ppc <pathtoMrbayes>/mb_mpi
> datafile.nex

Matt,
 It is possible that mpirun is launching two ranks on the same node
because only 3 nodes are available, causing duplicated processing at one
node (This assumes you are launching mpirun from one of the 4 nodes listed
in the machines file, and that your input files are on local directories.
Otherwise, please ignore this possibility.)

 If that assumption is true, you can avoid this by either
 1. Invoking mpirun from a machine that is not listed in your
machines file, or
 2. Adding one more nodes to your machines file. Note that you can list
all your nodes in your machines machines file without hurting yourself.
"-np 4" will cause MPI to use only four of them.

hope it helps,
Ecap

> I have it set up so that I am using specific machine files to designate
> which processors to run on. we run only 4 chains so i am only picking 4
> processors. I have not yet implemented it into the SGE yet.
>
> 2. When I include the quit command at the end of the MrBayes block I
> get a debugging error WARNING: Memory (2) has not been freed hit any
> key. This error makes it so the program does not quit automatically.
>
> Any ideas would be much appreciated.
>
> Matt
>
> _______________________________________________
> Bioclusters maillist  -  Bioclusters at bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters
>

 Srigurunath Chakravarthi (E-cap)
 MPI Software Technology India
 Parallel Computing solutions that few can parallel.(SM)