maker_2.10 hangs after completion with mpich2

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

maker_2.10 hangs after completion with mpich2

Greg Taylor
Hi,
   I've been running Maker_2.10 in a parallel environment using  
mpich2.  The job seems to run to completion, but then hangs, and I  
need to manually kill it. The problem seems to be with the script  
maintain.pl not recognizing that all jobs are complete. Have you seen  
this behaviour before, and do you have any suggestions?

thank you,
Greg Taylor

_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Problem: Low CPU usage with mpich2

Felix Bemm
Dear all,

I have another mpich2 related question. I have set up the mpich2
environment without the mpd daemon. So far it works for several programs
like clustal and blast. Unfortunately I am experiencing some problems
with maker. It seems that maker-mpi is not using more than one node
after some computational tasks. Moreover the whole annotation process
needs double the time than on a single node with 4 cpus. A short look on
nodes that idle reveal a maker process that is only using 3-4 % of the
cpu... And I have the feeling that even this cpu usage is only caused by
some write and read cmd's and not some computational ones. Maybe someone
can help me here!

Besides: Great fan of maker!

Best regards
Felix

--
Felix Bemm
Department of Bioinformatics
University of Würzburg, Germany
Tel: +49 931 - 31 83696
Fax: +49 931 - 31 84552
[hidden email]

_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Re: Problem: Low CPU usage with mpich2

Carson Hinton Holt
Re: [maker-devel] Problem: Low CPU usage with mpich2 Make sure you update to MAKER 2.11.  There are still some minor issues with using hydra rather than mpd.  Here is a graph of throughput using MPICH@ and hydra.

FYI: in 2.11, one node (the root node) is always dedicated to just data distribution, this is because trying to have the node also do processing in a thread can cause performance issues in perl (no shared memory for threads).  So a small number of nodes on single contig really slowed things down.  2.12 will be out within the next few days and has other improvements for stability with hydra.



--Carson




On 6/24/11 1:58 AM, "Felix Bemm" <felix.bemm@...> wrote:

Dear all,

I have another mpich2 related question. I have set up the mpich2
environment without the mpd daemon. So far it works for several programs
like clustal and blast. Unfortunately I am experiencing some problems
with maker. It seems that maker-mpi is not using more than one node
after some computational tasks. Moreover the whole annotation process
needs double the time than on a single node with 4 cpus. A short look on
nodes that idle reveal a maker process that is only using 3-4 % of the
cpu... And I have the feeling that even this cpu usage is only caused by
some write and read cmd's and not some computational ones. Maybe someone
can help me here!

Besides: Great fan of maker!

Best regards
Felix

--
Felix Bemm
Department of Bioinformatics
University of Würzburg, Germany
Tel: +49 931 - 31 83696
Fax: +49 931 - 31 84552
felix.bemm@...

_______________________________________________
maker-devel mailing list
maker-devel@...
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org



Carson Holt
Graduate Student
Yandell Lab
http:/www.yandell-lab.org/
Eccles Institute of Human Genetics
University of Utah

_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Re: maker_2.10 hangs after completion with mpich2

Carson Hinton Holt
In reply to this post by Greg Taylor
Re: [maker-devel] maker_2.10 hangs after completion with mpich2 MPICH2 recently switched the daemon from mpd to hydra.  Hydra does not clean up processes like it should.

Make sure you update to MAKER 2.11.  Within the next few days 2.12 will be released which has other improvements for hydra.  The issues with hydra are related to the fact MAKER is written in Perl.  Virtually all other MPI programs are written in C.  Perl does not respond to signals immediately as happens with C, rather it uses delayed safe signaling by default.  Unfortunately hydra does not handle no-immediate signal responses very well.  You can also use the MPICH2 mpd daemon instead of hydra.

--Carson




On 6/23/11 5:06 PM, "Greg Taylor" <gtaylor@...> wrote:

Hi,
   I've been running Maker_2.10 in a parallel environment using
mpich2.  The job seems to run to completion, but then hangs, and I
need to manually kill it. The problem seems to be with the script
maintain.pl not recognizing that all jobs are complete. Have you seen
this behaviour before, and do you have any suggestions?

thank you,
Greg Taylor

_______________________________________________
maker-devel mailing list
maker-devel@...
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org



Carson Holt
Graduate Student
Yandell Lab
http:/www.yandell-lab.org/
Eccles Institute of Human Genetics
University of Utah

_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Compiling Problems

Felix Bemm
In reply to this post by Greg Taylor
Hi,

so far I have figured out that my poor performance during usage of
mpich2 is related to not using hydra and v2.10 of maker. I am trying to
compile v2.11 and getting these warnings:

WARNING: the following files are missing in your kit:
         bin/mpi_evaluator
         bin/mpi_iprscan
         bin/mpi_maker
         lib/Parallel/MPIcar.c
         lib/Parallel/MPIcar.o
         lib/Parallel/MPIcar.pm
         lib/Parallel/MPIcar.xs
         lib/Parallel/typemap
         lib/Parallel/utils.c
Please inform the author.

Should I be worried about that?

Best regards
Felix

--
Felix Bemm
Department of Bioinformatics
University of Würzburg, Germany
Tel: +49 931 - 31 83696
Fax: +49 931 - 31 84552
[hidden email]

_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org
Reply | Threaded
Open this post in threaded view
|

Re: Compiling Problems

Carson Hinton Holt
Re: [maker-devel] Compiling Problems No.  You can ignore that.  It is just checking for files that aren’t supposed to be there anyway.  I have fixed this in what wil become the next version.

Thanks,
Carson


On 6/28/11 1:26 AM, "Felix Bemm" <felix.bemm@...> wrote:

Hi,

so far I have figured out that my poor performance during usage of
mpich2 is related to not using hydra and v2.10 of maker. I am trying to
compile v2.11 and getting these warnings:

WARNING: the following files are missing in your kit:
         bin/mpi_evaluator
         bin/mpi_iprscan
         bin/mpi_maker
         lib/Parallel/MPIcar.c
         lib/Parallel/MPIcar.o
         lib/Parallel/MPIcar.pm
         lib/Parallel/MPIcar.xs
         lib/Parallel/typemap
         lib/Parallel/utils.c
Please inform the author.

Should I be worried about that?

Best regards
Felix

--
Felix Bemm
Department of Bioinformatics
University of Würzburg, Germany
Tel: +49 931 - 31 83696
Fax: +49 931 - 31 84552
felix.bemm@...

_______________________________________________
maker-devel mailing list
maker-devel@...
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org



Carson Holt
Graduate Student
Yandell Lab
http:/www.yandell-lab.org/
Eccles Institute of Human Genetics
University of Utah

_______________________________________________
maker-devel mailing list
[hidden email]
http://box290.bluehost.com/mailman/listinfo/maker-devel_yandell-lab.org