Issues with Jacket HPC and License Server

[Old posts from the commercial version of ArrayFire] Issues and comments for download and installation. Getting up and running.

Moderators: melonakos, pavanky

Issues with Jacket HPC and License Server

Postby Berns » Thu Jun 28, 2012 11:23 am

Hi all,

we are trying to set up the new Jacket 2.2 on our Jacket HPC Cluster.
The problem seems to be that jacket seems to think we have a standalone license and therefore only runs on the last GPU of each node.

Here some additional info that might be helpful:
---------------------------------------------------------------------------------------------------------------------------------------------------
When I type gver on the headnode I get (alongside MATLAB license info):
-------------------------------------------------------------------------------------
AccelerEyes Jacket Version 2.2 (Build 77be88c)
-------------------------------------------------------------------------------------
<unfinished>

The unfinished part seems strange, but even stranger that jacket works fine on one GPU of each node.
When i type ginfo i get:
Jacket v2.2 (build 77be88c) by AccelerEyes (64-bit Linux)
License: Standalone (/usr/local/jacket/engine/jlicense.dat)
CUDA toolkit 4.2, driver 4.2 (295.41)
GPU1 Tesla M2090, 5376 MB, Compute 2.0 (single,double) (in use)
GPU2 Tesla M2090, 5376 MB, Compute 2.0 (single,double)
GPU3 Tesla M2090, 5376 MB, Compute 2.0 (single,double)
GPU4 Tesla M2090, 5376 MB, Compute 2.0 (single,double)

>> gselect(2)
Warning: 4 GPUs available. Jacket MGL license not available. Using GPU1 only.

This might indicate a wrong license file, but when i look at the indicated file, I see something very reasonable:
SERVER (...) 28000
VENDOR jacket
FEATURE jacket jacket 2.2 permanent 28 SIGN=(...)
FEATURE hpc jacket 2.2 permenent 28 SIGN=(...)

I did not want to copy the whole file here as i am not sure whether I am allowed to by my institute or even accelereyes. Could send it somehow via an encrypted mail or sth. if needed.
In addition: Licenses are checked out when i issue a ginfo command on any of the nodes:
(jacket) OUT: "hpc" xxx@xxx

And also only one GPU is used when invoking spmd, parfor or the scheduler we usually use.
Output from nvidia-smi shows all running on the same GPU when putting load onto cluster ( last one):
As a last note: using setenv FORCE_MGL 1or HPC 1 does not help either as indicated somewhere on the website, no error anymore when using gselect, but no change in GPU currently in use either.

Any help would be deeply appreciated. Kinda urgent too. xD So, any ideas how we could solve this/what is wrong?
Berns
 
Posts: 1
Joined: Thu Nov 24, 2011 10:41 am

Re: Issues with Jacket HPC and License Server

Postby vishy » Thu Jun 28, 2012 1:03 pm

Hi Berns,

This appears to be a problem that requires us to take a longer look. Can you email support@accelereyes.com, with the details of this problem and all information described at http://www.accelereyes.com/support/technical_support?
--------------------------------------------------
Vish Venugopalakrishnan
Software Engineer (Q/A)
AccelerEyes LLC
vishy.v@accelereyes.com

--------------------------------------------------
Resources:
Getting Started, FAQ, Tips, Syntax
User avatar
vishy
 
Posts: 411
Joined: Thu Apr 16, 2009 11:46 am


Return to [archive-commercial] Download & Installation

cron