Since we’re running MPI jobs, we allocate entire nodes (i.e. partitions are set 
up as OverSubscribe=Exclusive) to avoid resource contention.  So most of our 
scripts look like this:

#SBATCH --ntasks=288
#SBATCH --constraint=[wes|san|has|bro]

And each job gets enough nodes to satisfy the task count.  This sounds a little 
different from your workflows.

> sorry to hijack, but we've been considering a similar configuration, but I was
> wondering what happens if you don't set a processor type?   Will it scatter 
> across types?

As Paul said, yes, you will get a mix if you don’t specify anything.  We deal 
with this by using job_submit.lua to require users to request at least one type.

From: slurm-users <slurm-users-boun...@lists.schedmd.com> on behalf of Kevin 
Van Workum <v...@sabalcore.com>
Reply-To: Slurm User Community List <slurm-users@lists.schedmd.com>
Date: Wednesday, June 20, 2018 at 1:15 PM
To: Slurm User Community List <slurm-users@lists.schedmd.com>
Subject: Re: [slurm-users] Multiple job constraints



On Wed, Jun 20, 2018 at 10:29 AM, Vicker, Darby (JSC-EG311) 
<darby.vicke...@nasa.gov<mailto:darby.vicke...@nasa.gov>> wrote:
Yes, Boolean constraints do work with minimal configuration. We run mostly MPI 
jobs on our cluster and want our jobs to run on a single processor type. We 
assign the processor types as a feature and then the sbatch requests 
—constraint=[wes|san|has|bro] to run on whatever processor type is free. Works 
great.

Can you also request a different job layout for each processor type? As in the 
example I gave?


On Jun 20, 2018, at 9:15 AM, Kevin Van Workum 
<v...@sabalcore.com<mailto:v...@sabalcore.com>> wrote:
Hello,
We currently use an in-house customized version of Torque and an in-house 
scheduler. I'm now considering Slurm instead. We have a feature in our 
scheduler where by users can submit jobs with multiple-prioritized job 
specifications (constraints). For example, a job can request 4 nodes of type X 
with 16 cores per node OR 2 nodes of type Y with 20 cores per node. The 
scheduler will prefer the first job spec but may use the second job spec if the 
first can not be satisfied within certain QoS limits.
I saw some ability to use boolean constraints in the docs for Slurm, but would 
like to know if this behavior is possible out of the box.
Thanks,
Kevin

--
Kevin Van Workum, PhD
Sabalcore Computing Inc.
"Where Data Becomes Discovery"
http://www.sabalcore.com
877-492-8027 ext. 1011

[Image removed by sender.]



--
Kevin Van Workum, PhD
Sabalcore Computing Inc.
"Where Data Becomes Discovery"
http://www.sabalcore.com
877-492-8027 ext. 1011

[Image removed by sender.]

Reply via email to