SLURM Workload Manager

From HPC users
Jump to navigationJump to search

Slurm is an open source cluster management and job scheduling system. With Slurm we can send, queue, modify, and inspect jobs and also have the possibility to manage our resources, e.g. creating a reservations for a specific time range.

Reservation

e.g. for specific courses, it can be mandatory to keep some resources free for a specific time span.

Setting up a reservation

scontrol create reservation reservationname=moleco partition=mpcb.p nodecnt=5 account=kurskonto start=2019-03-01T08:00 duration=12:00:00 flags=DAILY

In this example, we set the following options:

  • reservationname=moleco
    • The reservation needs a name so that the users can send their jobs to the corresponding reserved nodes.
  • partition=mpcb.p
    • We need to set a partition, ideally one that is not working on full capacity.
  • nodecnt=5
    • Identify number of nodes to be reserved.
  • account=kurskonto
    • Only the specified account/group will be able to use the reserved resources.
  • start=2019-03-01T08:00
    • Set the date and time to start the reservation.
  • duration=12:00:00
    • For this course, the reservation will only be effective on daytime, 08:00 AM + 12h.
  • flags=DAILY
    • The reservation will be active on a daily basis.

In this example, the reservation has to be cancelled manually. Alternatively, we could set a value for EndTime, but in that case the reservation would be kept up at night, since the option duration collides with EndTime.


Working on existing reservations

To get an overview of the existing reservations, type in

scontrol show reservation

To update the values for an existing reservation, type in

scontrol update reservation <option=value>