<html>
<head>
<meta content="text/html; charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
Hello<br>
<br>
I'm trying to get maker to run on MPI using several nodes. I have
an installation set up by a colleague which includes maker 2.27 and
openmpi-1.4.3. <br>
<br>
Previously it has only been used (here at EBI) with maker processes
running on one node only, but i find that it can wait a very long
time before being scheduled by LSF.<br>
<br>
The command used to submit is like this (as per recommendations from
systems) (uses 8 cpus on each of 8 nodes)<br>
<br>
<div class="container" title="Hint: double-click to select code">
<div class="line number1 index0 alt2"><code class="java plain">export
OMP_NUM_THREADS=</code><code class="java value">64</code></div>
<div class="line number2 index1 alt1"><code class="java plain">bsub
-q mpi -M </code><code class="java value">40000</code> <code
class="java plain">-R </code><code class="java string">"rusage[mem=40000]
&& span[ptile=8]"</code> <code class="java plain">-n
</code><code class="java value">64</code> <code class="java
plain">-o lsf_log -a openmpi mpirun.lsf -np </code><code
class="java value">64</code> <code class="java plain">-mca
btl tcp,self maker </code><code class="java value">2</code><code
class="java plain">>&</code><code class="java value">1</code></div>
</div>
<br>
and requires environment be set up in ~/.bashrc for openMPI.<br>
<br>
This runs but produces a lot of errors like:<br>
<br>
DBD::SQLite::db do failed: database is locked at
/nfs/production/panda/ensemblgenomes/external/maker/2.27_mpi/maker/bin/../lib/GFFDB.pm
line 407.<br>
<br>
<br>
I've looked at
<a class="moz-txt-link-freetext" href="https://groups.google.com/forum/#!searchin/maker-devel/database$20locked/maker-devel/TscBgbQfBX4/pae016DqlIMJ">https://groups.google.com/forum/#!searchin/maker-devel/database$20locked/maker-devel/TscBgbQfBX4/pae016DqlIMJ</a>
which suggests that "It means that your GFF3 results will not be
integrated" (but i'm not sure what's meant by that, but the number
of genes i'm getting is around 2k, expect more like 15k) and that
the problem is SQLite using NFS (a known issue), and the fix is to
use /tmp.<br>
<br>
I have TMP= set as per default in maker_opts.ctl, and there are
maker directories in /tmp on the runtime nodes, but the database (i
guess) is in
/nfs/...../maker/<maker_out_dir>/<prefix>.scf.db.<br>
<br>
I don't see how i could set the working directory to a non-NFS file
systems and still use more than one node, but this error only seems
to appear (so far) with est2genome, not when running SNAP/ Augustus.<br>
<br>
Is there a work around to stop getting the locked error or some way
to recover from it after maker has finished? Or is it necessary to
run the est2genome step (or maker generally) on one node? <br>
<br>
An obvious option is to split the assembly but i was hoping to avoid
that.<br>
<pre class="moz-signature" cols="72">--
malcolm hinsley | EnsEMBL Genomes | +44 (0)1223 49 4669
European Bioinformatics Institute (EMBL-EBI)
European Molecular Biology Laboratory
Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD
United Kingdom </pre>
</body>
</html>