[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] How to specify slot specific information into same machine class ad name?



Hi Carsten,

I am not sure, if I got you correctly, but maybe concurrency limits [1] might be an option? I.e., to define a consumable resource per your fragmented slots - classical example might be a limited number of licenses for an application, that can be used in parallel. If each of your slots gets a concurrency limit assigned, that can be "used up" by the users.

Cheers,
  Thomas



[1]
https://htcondor.readthedocs.io/en/latest/admin-manual/cm-configuration.html#concurrency-limits

On 11/02/2025 12.22, Carsten Aulbert wrote:
Hi all,

for a new project, we are currently defining four slots per execute node to ensure GPU/CPU/NVMe usage does not cross inter-CPU boundaries, e.g.

ENFORCE_CPU_AFFINITY = True

NUM_SLOTS_TYPE_1ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ = 1
SLOT_TYPE_1ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ @=slot
 Âcpus=12
 Âram=20%
 Âswap=0%
 ÂGPUS = 1 : DevicePciBusId == "0000:2A:00.0"
@slot
SLOT_TYPE_1_PARTITIONABLEÂÂÂÂÂÂÂ = True
SLOT1_CPU_AFFINITY = 0,2,4,6,8,10,24,26,28,30,32,34
[...]

and we establish a fixed mapping between each such slots and a locally available data set which we simply "number" from 00..58.

We now want to create and start a DAG which contains a single job for each of these possible numbers and should user `requirement` to match to the proper slot on the proper target machine.

However, I'm not sure which way to achieve this easily.

First stop was

IDX1 = 10
IDX2 = 45
SLOT1_STARTD_ATTRS = IDX1
SLOT2_STARTD_ATTRS = IDX2

but of course this would inject IDX1 into slot1 and IDX2 into slot 2 which would make the requirement line somewhat lengthy (testing if any of IDX1/2/3/4 matches the wanted number).

I've not yet tried to use some logical argument for IDX as I'm not sure in which context it would be evaluated, if at all.

I don't think I could inject per slot values via STARTD_CRON, so how could I approach this?

Cheers

Carsten


_______________________________________________
HTCondor-users mailing list
To unsubscribe, send a message to htcondor-users-request@xxxxxxxxxxx with a
subject: Unsubscribe

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

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature