Mailing List Archives
Public Access
|
|
|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[HTCondor-users] -better-analyze oddity
- Date: Wed, 30 Dec 2015 18:18:04 -0500
- From: Michael Di Domenico <mdidomenico4@xxxxxxxxx>
- Subject: [HTCondor-users] -better-analyze oddity
Suppose you have a server with 8 cpus and 4 gpus, with a single
dynamic slot, cpus=8,gpus=4,mem=auto
user A starts eight jobs on the cpus, but does not request any gpus
user B starts four jobs on the gpus
user B's jobs will not start until user a vacates a slot, preemption is off
however, when user B runs 'condor_q -better-analyze <job>' the output
indicates that there are Zero gpu slots available in the pool in which
to run user B's job. which is understandable since the dynamic slot
provisioned for user A did not include a gpu
however, to the untrained, this makes it appear that there are Zero gpus period.
In my instance user B's jobs did run just fine when user A vacated,
but is there any way to smarten up better-analyze or change my config
around so that if user B's jobs sit idle it's more clear that their
waiting on a slot to be provisioned with a gpu. the current output
makes it seem like removing the gpu from their job is the only way the
job will run