<[email protected]> writes:

> Following up on the suggestion from from William Hay about using a
> JSV to "wildcard" a PE request in order to deal with MPI jobs on a
> multi-architecture cluster,

What exactly was the suggestion?  I don't think I saw it, and I wonder
why it needs extending from the recipe I've posted before.

> I've found that the JSV unexpectedly alters
> a resource request from booleans to float.
>
> In ~/.sge_request, I've got:
>       -l centos5
>
> for a job submitted without the JSV, this works fine and sets a hard resource
> request of "centos5=TRUE".
>
> However, when I submit the job and call the JSV (qsub -jsv /path/to/myjsv 
> myjob),
> the value of "centos5" in the hard resource list is changed to "0.000000".

Fixed in SGE 8.0.0c (imported from Univa).

> If I change the line in ~/.sge_request to read:
>
>       -l centos5=TRUE
>
> then it is not altered by the JSV. We're using SGE 6.2u5.
>
> The JSV consists of:
>
> #############################
> #!/bin/sh
> # from
> #   http://www.mentby.com/Group/grid-engine/default-resource-per-queue.html
>
> # JSV to convert PE names to wildcard version, in order to allow users to 
> submit jobs
> # with:
> #   -l pe openmpi
> # and have it transparently refer to
> #   -l pe openmpi-Intel*
> # or
> #   -l pe openmpi-AMD*
>
> jsv_on_start() {
>       return
> }
>
> jsv_on_verify() {
>       pe=$(jsv_get_param pe_name)
>       case "$pe" in
>           openmpi )
>             jsv_set_param pe_name "$pe-*"
>             jsv_sub_add_param l_hard openmpi TRUE   # in case it wasn't set 
> already

Why is that complex needed?

-- 
Community Grid Engine:  http://arc.liv.ac.uk/SGE/
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to