On 15/12/15 12:11, Simpson Lachlan wrote:

> Hi,

Hi Simpson,

> I think of SLURM - perhaps incorrectly - of consisting of a head node
> and a collection of worker nodes.

That's right, though some people call the head node the management node
and have a separate login node for users.

> Installing for the first time, I would like to get some clarification
> please.
> 
> On the head node, I run slurmctld. On each worker node I run slurmd,
> but not slurmctld.

That's right.

> I only run slurmd on the head node if I would also like the head node
> to be a worker node? If this is the case, then I also need to add the
> head node's hostname to the NodeName and PartitionName vars in the
> slurm.conf.

Got it.

> If I don't want the head node to run as a worker, then there is no
> need for me to add it to either NodeName or PartitionName in
> slurm.conf.
>
> Have I got that right?

Indeed - you seem to be heading down the right track.

> From there, I presume that a simple way to set up the system would be
> to run the logging/accounting db (the slurmdbd node?) on the head
> node?

You can do that yes.

> Is this a sane set up or is combining the slurmctld and slurmdbd
> processes on a single node a bad idea?

We only do that on a test cluster, but I'm just helping another group
bring up a Slurm cluster and that's the current plan there.

> Finally, in what way does a login node differ from a head, worker or
> accounting node, and is there anything special that I need to do to
> that node - apart from implementing some sort of authorization
> system?

The key is for that is Munge - you need to make sure you've got that
configured & running everywhere (same key, correct permissions, starting
on boot).

Best of luck!
Chris (also in Melbourne)
-- 
 Christopher Samuel        Senior Systems Administrator
 VLSCI - Victorian Life Sciences Computation Initiative
 Email: [email protected] Phone: +61 (0)3 903 55545
 http://www.vlsci.org.au/      http://twitter.com/vlsci

Reply via email to