[Iplant-api-dev] GeneSeqer-MPI refactored for Agave

[Roger Barthelson]

Hi Nirav-

Yes, I was about to post this on api-dev. I had more time to focus on it this week, and I was able to confirm that I can run a GeneSeqer job successfully when I submit directly to the queue, eg. one with the setting -N 4 -n 64. When I submit through Agave, either via the DE or with the CLI, I get a job script that sets up the job as -N 4 -n 16. I have tried adding a line to the jason with this custom directive:

"customDirectives": " -N 4 -n 64”, 

But it still gives me just 16 threads. I don’t know if I’m missing something. I think this is the first Agave app I’ve built that uses multiple nodes. My understanding of the documentation is that when you submit a job as -N 4 -n 64, you are telling the system you are using 4 nodes and running a total of 64 mpi tasks. Is my understanding incorrect? Is there a special setting to put in an Agave app’s json to designate this kind of job? Is there a conflict with designating the system as 16 processors per node? 

Roger
-- 
Roger A Barthelson, PhD
iPlant Collaborative
BIO5/University of Arizona
rogerab at email.arizona.edu
520-977-5249

Sent with Airmail

On February 24, 2015 at 3:12:26 PM, Nirav Merchant (nirav at email.arizona.edu) wrote:

Hi Roger,
I wanted to revisit the challenges we have had with getting the GeneSeqer MPI version migrated from foundation API to Agave. My understanding is that the bulk of the work is completed but the part that utilizes the multiple nodes (during MPI) is not being honored during execution.

I am including Matt on the thread so he can provide some pointers for creative ways forward. Please articulate the issue and I am sure we can find suitable options

Regards,
Nirav

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.iplantcollaborative.org/pipermail/iplant-api-dev/attachments/20150224/d856c9ac/attachment.html