Ticket 3183 - srun to send ntasks_per_core/socket option to slurmctld for step scheduling
Summary: srun to send ntasks_per_core/socket option to slurmctld for step scheduling
Status: CONFIRMED
Alias: None
Product: Slurm
Classification: Unclassified
Component: Scheduling (show other tickets)
Version: 17.02.x
Hardware: Linux Linux
: 5 - Enhancement
Assignee: Unassigned Developer
QA Contact:
URL:
Depends on:
Blocks:
 
Reported: 2016-10-17 09:17 MDT by Moe Jette
Modified: 2019-02-21 02:14 MST (History)
2 users (show)

See Also:
Site: SchedMD
Slinky Site: ---
Alineos Sites: ---
Atos/Eviden Sites: ---
Confidential Site: ---
Coreweave sites: ---
Cray Sites: ---
DS9 clusters: ---
Google sites: ---
HPCnow Sites: ---
HPE Sites: ---
IBM Sites: ---
NOAA SIte: ---
NoveTech Sites: ---
Nvidia HWinf-CS Sites: ---
OCF Sites: ---
Recursion Pharma Sites: ---
SFW Sites: ---
SNIC sites: ---
Tzag Elita Sites: ---
Linux Distro: ---
Machine Name:
CLE Version:
Version Fixed:
Target Release: ---
DevPrio: ---
Emory-Cloud Sites: ---

Attachments
Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this ticket.
Description Moe Jette 2016-10-17 09:17:02 MDT 
The job step can not over-ride --ntasks-per-core or --ntasks-per-socket parameters set by the job allocation
Comment 1 Alejandro Sanchez 2016-10-17 09:25:14 MDT 
Shouldn't --ntasks-per-node be included in the same change pack too?
Comment 2 Moe Jette 2016-10-17 09:35:10 MDT 
(In reply to Alejandro Sanchez from comment #1)
> Shouldn't --ntasks-per-node be included in the same change pack too?

The code would need to be examined to determine which options should be added and I have not done that yet. I have only determined that --ntasks-per-core and --ntasks-per-socket parameters are not in the step create request.
Comment 4 Zhengji Zhao 2016-10-21 12:55:01 MDT 
Dear Moe,

While you make the --ntasks-per-core option to work on the job step creation time (I mean in the srun command line after the nodes are alreay allocated to the job), the --ntasks-per-socket is also very useful for us if you make it work at job step creation time. 

Thanks,
Zhengji
Comment 5 Moe Jette 2016-10-21 14:53:30 MDT 
(In reply to Zhengji Zhao from comment #4)
> Dear Moe,
> 
> While you make the --ntasks-per-core option to work on the job step creation
> time (I mean in the srun command line after the nodes are alreay allocated
> to the job), the --ntasks-per-socket is also very useful for us if you make
> it work at job step creation time. 

The earliest this could be available is in Slurm version 17.02, to be released in February.
Comment 6 Zhengji Zhao 2016-10-21 15:05:14 MDT 
Thanks a lot for the info!

Zhengji

On Fri, Oct 21, 2016 at 1:53 PM, <bugs@schedmd.com> wrote:

> *Comment # 5 <https://bugs.schedmd.com/show_bug.cgi?id=3183#c5> on bug
> 3183 <https://bugs.schedmd.com/show_bug.cgi?id=3183> from Moe Jette
> <jette@schedmd.com> *
>
> (In reply to Zhengji Zhao from comment #4 <https://bugs.schedmd.com/show_bug.cgi?id=3183#c4>)> Dear Moe,
> >
> > While you make the --ntasks-per-core option to work on the job step creation
> > time (I mean in the srun command line after the nodes are alreay allocated
> > to the job), the --ntasks-per-socket is also very useful for us if you make
> > it work at job step creation time.
>
> The earliest this could be available is in Slurm version 17.02, to be released
> in February.
>
> ------------------------------
> You are receiving this mail because:
>
>    - You are on the CC list for the bug.
>
>
Comment 7 Danny Auble 2016-11-08 14:34:47 MST 
Zhengji,

There is code in 17.02 now that handles --ntasks-per-core --ntasks-per-socket much better than it did previously for steps.

The options for the step still don't work though as requested as the allocation has already been granted at that point so you would still need to specify the option at allocation time, but it is much better as the task layout for the step now takes into consideration the options as given in the allocation.  (See bug 926)

It would be quite involved to add this to work on a step differently than was requested on the allocation.  Hopefully what has been done for 926 will give you close to what you are looking for.