No start time listed in squeue

HPC Support
1

The past few weeks I have noticed that start my start times are almost nonexistent and might not show up for over a day. I usually use the start time to consider whether or not I will start another batch of jobs immediately or work on something else in the mean time. For example, my latest array job:

squeue --format="%.18i %.9P %.40j %S %.8T %.10M %.9l %.6D %R" -u $USER
             JOBID PARTITION                                     NAME START_TIME    STATE       TIME TIME_LIMI  NODES NODELIST(REASON)
   14069541_[1-16] public-cp                              BEAST_rerun N/A  PENDING       0:00   4:00:00      1 (Priority)

I am curious if there is something I am doing that is preventing my jobs from running? Or is it just that my priority is so low that the job scheduler is uncertain of the start time?

I know I’ve asked for and submitted quite a few jobs lately, so I wonder if the later is the case.

Thanks in advance!

(PS - I know asking for a full four days is not encouraged, but these beast runs can take quite a while to run)

2

Dear @Cody.Cardenas

That is probably the reason. In fact, it is not your priority that is low, it is that you are using the public-cpu partition, which has only 8 compute nodes.

Is it possible to use some checkpointing (restart) perhaps?

Another suggestion is to use Bamboo: 29 compute nodes (128 cores each) are sitting idle waiting for your job. Bonus, the home storage is on SSD and super fast.

3

Ive not had access to bamboo since it was live, I’ll look into how to access it.

I didn’t think that Beast had checkpointing, but I’ll have a look into it and reconsider how to format my jobs so I can meet the 100-200 million MCMC chain iterations necessary with less demanding times.

Thanks for the feedback @Yann.Sagon!

4

Every user has access to the three clusters:

https://doc.eresearch.unige.ch/hpc/access_the_hpc_clusters#cluster_connection

5

A post was split to a new topic: Use Beast checkpointing with slurm job array