Our node hardware configuration is as follows:
CPU Cores: 20
Memory: 512GB
GPUs: 8
With the following srun command:
srun --gres=gpu:2 --pty bash
We are trying to set Slurm policies for the following conditions:
Memory: User gets 32GB (Memory/number of GPUs on node) times number of GPUs
Ok, never mind, I think I got it, it seems that scontrol reconfigure was
not enough, I also had to restart the slurm daemons and it seems to work
now.
Regards!
El 23/05/18 a las 13:00, Antonio Lara escribió:
Hi again,
Also, if I write the nodes names wrong in topology.conf (that is,
nodes
Hi,
I have been trying out using 'parent' as a value for Fairshare and
'sshare -a' shows me the following:
Account User RawShares NormSharesRawUsage
EffectvUsage FairShare
-- -- --- ---
- --
ro
Hi,
Does anyone know whether one can modify existing account information by
making a dump, deleting everything apart from the users to be modified,
modifying those uses to get, say
Cluster - 'tux':DefaultQOS='long':Fairshare=1:QOS='long,medium,short'
Parent - 'root'
Account - 'dept01':Descr
Hi again,
Also, if I write the nodes names wrong in topology.conf (that is, nodes
that are not specified in slurm.conf, in the line describing the
partition, under "Nodes="), when I do scontrol reconfigure, there is no
complaint or messages in the logs, so it seems like the topology.conf
file
Hello,
I'm trying to use the topology/tree plugin to isolate nodes in different
"groups", so that jobs can be allocated only on nodes belonging to one
such group, and not in nodes from other groups. I think I'm missing
something, because Slurm doesn't seem to take this topology into
considera
Lev Lafayette writes:
> On Tue, 2018-05-22 at 13:56 +0200, Bjørn-Helge Mevik wrote:
>>
>> (including the newlines!). Our scripts didn't expect newlines in job
>> names, and so became angry. :)
>>
>
> That's beautiful in its ugliness and is going in my introductory
> training material. :)
:D
