Re: [HTCondor-users] Dynamic slots and limits by job profiles

[Jason Patton <jpatton@xxxxxxxxxxx>]

Hi Charles,

In our local pool, we give each machine one "IoHeavy" resource and ask users whose jobs are disk IO heavy to use "request_ioheavy = 1" in their jobs. Here's a post from the archive with more details:

https://www-auth.cs.wisc.edu/lists/htcondor-users/2022-September/msg00013.shtml

Maybe you could do something similar, where you give your machines a number of "simulation resources" and "rendering resources" depending on the machine's total resources, and then have your users request these resources for these types of jobs? E.g. to allow 10% of CPUs to be available at a time for simulation jobs and 20% of CPUs to be available at a time for rendering jobs:

MACHINE_RESOURCE_Simulation = INT(0.1 * $(DETECTED_CPUS))

MACHINE_RESOURCE_Rendering = INT(0.2 * $(DETECTED_CPUS))

Then jobs could specify either request_simulation = 1 or request_rendering = 1 (or you could even have submit transforms add RequestSimulation = 1 or RequestRendering = 1 to the jobs if you think you can classify these jobs in your access point config).

Jason

On Thu, Mar 23, 2023 at 5:27âAM Charles Goyard <cgoyard@xxxxxxx> wrote:

Hi all,

again I'm requesting some help !


On our HTCondor pool (v10.0), we have dynamic slots enabled.

Smaller machines can accept 1 job, medium ones accept 2 jobs, and the
higher-end computers accept up to 4 jobs.

Say our users run 3 types (or profiles) of jobs:

- Compositing
- Rendering
- Simulation

The jobs declare a property with the job type (in an env var for example).

Running multiple Compositing tasks at once does not cause a problem, but
having 4 Rendering or Simulation is not really great.

What I would like to achieve is to define an _expression_ on the execute
nodes that says :

- accept at most one Simulation job at once.
- accept at most two Rendering jobs at once.
- accept any number of Compositing jobs.

So for example a single execute node could run :

sim.  render comp. Âtotal
-----------------------------
1Â Â Â Â2Â Â Â Â1Â Â Â Â4
1Â Â Â Â1Â Â Â Â2Â Â Â Â4
1Â Â Â Â0Â Â Â Â3Â Â Â Â4
0Â Â Â Â2Â Â Â Â2Â Â Â Â4
0Â Â Â Â1Â Â Â Â3Â Â Â Â4
0Â Â Â Â0Â Â Â Â4Â Â Â Â4


Since there are only 6 cases, I'm ok this building a super-long
_expression_ :).

I found a recipe that looks like this kind of thing, but for static
slots here : https://htcondor-wiki.cs.wisc.edu/index.cgi/wiki?p=HowToReserveSlotForSpecialJobs .

But I can't get my head around on how to achieve this setup with dynamic
slots. How can I get information on the type of job running at a given
time on a execute node ? (this sounds a bit like my question about how
to count the number of dynamic slots discussed in December).

Thanks,

--
Charles
_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe
You can also unsubscribe by visiting
https://lists.cs.wisc.edu/mailman/listinfo/htcondor-users

The archives can be found at:
https://lists.cs.wisc.edu/archive/htcondor-users/