Matt,

To allow the jobs from each owner (11, 12, and 13) to be prioritized based on 
their usage, do not include the fairshare=parent setting.

Otherwise, your settings should result in usage that reflects the shares you 
have configured, assuming all three owners have been submitting enough jobs to 
the queue.  To check, run “sshare -a” from time to time.

Don

From: Guglielmi Matteo [mailto:[email protected]]
Sent: Monday, January 28, 2013 7:47 AM
To: slurm-dev
Subject: [slurm-dev] Independent FairShare Configuration for N Cluster Segments

I have slurm running on a confederation of clusters but at
slurm level, I've defined a single cluster.

For each physical cluster I've created an account and to each
account I've associated the corresponding "owner" users.

To each account I've also associated a single partition which
includes all the nodes of that particular cluster.


                           root
                             |
                             |
     +-----------+-----------+----------+-----------+
     |           |           |          |           |
  account1   account2    account3    account4    account5
     |           |           |          |           |
     + owner11   + owner21   +          +           + owner51
     + owner12   + owner22   ...        ...         + owner52
     ~           ~                                  ~
     + owner1N   + owner2K                          + owner5J


Because each owner user is associated to a single account, I
need to configure the faishare so that the priority for each
user belonging the same account is not affected by what's happening
to the accounts.

So, if I set:

fairshare=parent

for all 5 accounts and, for example:

fairshare=47  (owner11)

fairshare=13  (owner12)

fairshare=40  (owner13)

fairshare=0   (owner14,... owner1N)

do I get the corresponding shares of 47%, 13% and 40% for the
three owners of account1?

Are the corresponding shares independent from the load of experienced
by the other cluster segments (account2,... account5)?

Thanks,

--matt

<<inline: ~WRD000.jpg>>

Reply via email to