batch queue manager basic information

Post Reply
jwk
Posts: 81
Joined: Sat Jun 20, 2020 3:37 am
Full Name: John Keller
Organization: University of Alaska Fairbanks
Subdiscipline: Chemistry

batch queue manager basic information

Post by jwk »

I'm not understanding the batch queue manager (BQM) documentation for webmo 23 ent.
Starting way back at the beginning...
OK, so webmo is installed on a local server where I am the admin. And I can connect via ssh to a remote compute cluster running SLURM, and run gaussian jobs there under my account.
(1) To utilize webmo's BQM functions, what changes or installations would need to be done (by cluster admins) on the remote compute cluster?
(2) Which account on the cluster would run jobs sent from webmo?
(3) Can anyone with an account on the local webmo system run jobs on the remote cluster via BQM? Or does the webmo user choosing BQM also need an account on the remote compute cluster?
schmidt
Posts: 98
Joined: Sat May 30, 2020 3:00 pm
Full Name: JR Schmidt
Organization: WebMO, LLC

Re: batch queue manager basic information

Post by schmidt »

(1) Most critically, WebMO needs to be installed on the head node of the cluster that is running SLURM. In other words, WebMO should be installed on the head node of the cluster. The detailed requirements for integrating WebMO with a batch queue can be found here: https://www.webmo.net/link/help/BatchQueues.html But in short, these are almost always automatically satisfied by installing WebMO on the head node.

(2) Which account is used depends on how you setup / configure WebMO. In the simplest approach, all jobs are executed and queue as user 'webmo'. Alternatively, if you integrate with external authentication, WebMO can execute jobs under each individual users UID (assuming corresponding system accounts exist).

(3) This question is largely irrelevant due to (1). WebMO needs to be installed on the head node of cluster in order to use the queue. (Note that with some work, this is probably not strictly true. If (a) an NFS file system is shared between the systems and (b) the other system is a SLURM submit host, it may be possible to make it work. But there is little to no reason to use such an approach, unless you want to strictly separate the web server duties from the head node.
jwk
Posts: 81
Joined: Sat Jun 20, 2020 3:37 am
Full Name: John Keller
Organization: University of Alaska Fairbanks
Subdiscipline: Chemistry

Re: batch queue manager basic information

Post by jwk »

An additional "local" webmo installation is irrelevant: install webmo on the SLURM head node. Users connect directly to it, and the jobs are directed to SLURM as user "webmo".
Post Reply