Closed HenrikBengtsson closed 6 years ago
Good point, I think we can remove "bigmem"... Do you know how many bigmem and "smallmem" nodes are available e.g. for Costello lab?
Ok, let's plan on removing that.
smallmem
resource/node; a node is flagged as either bigmem
or not.pbsnodes | grep -E "(^n[0-9]+|big|small)"
on the cluster to see which nodes have the bigmem
flag. See https://ucsf-ti.github.io/tipcc-web/about/specs.html for which nodes Costello Lab has access to.I see the picture now... So we have 10 nodes with 32-64GB of RAM, which could be used for small memory jobs. Good to know! I will work on removing bigmem, I guess a sed one liner should do it...
Removed!
Related to this, @shuntsman-ucsf, did you manage to drop bigmem
using qalter
on TIPCC? If not, we can do a quick release of the current develop version where bigmem
is no longer used.
what is qalter
??
I did remove it from *.pbs files...
what is qalter ??
https://www.jlab.org/hpc/PBS/qalter.html - Scott gots lots of jobs already queued up and by using qalter
he can modify the existing resource specs of those jobs.
I did remove it from *.pbs files...
Yup, thxs for that - that's in the develop
branch. Scott is always running the latest stable release (to avoid him using a broken/poorly tested develop commit among other things).
Related to this, @shuntsman-ucsf, did you manage to drop
bigmem
usingqalter
on TIPCC? If not, we can do a quick release of the current develop version wherebigmem
is no longer used.
I did not see any mention of "bigmem" in any of the text output by 'qstat -f'. And when I released several modified jobs from hold, some went to a new node for the first time, so hopefully not an issue anymore. I only used the following commands:
qalter -l nodes=1:ppn=10 jobid
qalter -l vmem=20G jobid
Sounds like those qalter
calls of yours caused any bigmem
:s to be dropped. Good. Then we don't need to rush out a new release.
Several *.pbs scripts have PBS declarations that request
bigmem
nodes, e.g.Do we really need that? I would think specifying
mem/vmem
would take care of that. I don't know the history ofbigmem
but it could be a legacy from old times.Having
bigmem
in there could prevent a job from ending up on a non-bigmem
nodes if one would submit the PBS script with sayqsub -l nodes=1:ppn=6,vmem=50gb ...