mothur MPI at CLUMEQ

Theory behind mothur
1

Hi,

The CLUMEQ (http://www.clumeq.ca/), a research consortium for high performance computing in Québec, Canada, currently has new users who wish to use our supercomputers to accelerate the compute time of mothur.

I am providing technical support to some of these users on one our machine, and I have some questions regarding mothur-MPI.

  1. A list of commands compatible with MPI was listed in [1], is it still valid? Have you added some commands to this list since?
  2. In [2], it is reported that the MPI version of mothur is slower. Has mothur been tried on a supercomputer? Do you have speedup results? Are future developments with MPI on mothur roadmap?
  3. Would it be possible to contribute to the development of the computation distribution aspect of mothur?

Regards,

PS: Good job on the spambot filter for you forum. The question is quite tricky for a computer engineer, I almost came to the conclusion that I might be a spambot ;).

[1] multiple processors and mpi
[2] Running mothur in mpi

2
  1. A list of commands compatible with MPI was listed in [1], is it still valid? Have you added some commands to this list since?

Hmmm, not sure - any command that uses the “processors” option will work with MPI.


2. In [2], it is reported that the MPI version of mothur is slower. Has mothur been tried on a supercomputer? Do you have speedup results? Are future developments with MPI on mothur roadmap?

Yup. We don’t really see a speed up over what we see using the processors option without using MPI. We probably won’t be doing too much more with this. Of the things that can be parallelized, the limitation is more memory than speed.



3. Would it be possible to contribute to the development of the computation distribution aspect of mothur?

Go for it. The code is up on GitHub: https://github.com/mothur/mothur

3

Are you guys working on this ? because this summer may be i will need to run mothur on Clumeq…

for the moment im having a nice results with my 2 processors on my laptop !

…if the code could be adapted to run on clumeq-parallel … it could be sexy !