Limit max nodes per group


#1

how to cap the usage of each group to a maximum of 7 nodes in the cluster


#2

Hi team

please help me on this issue.

Thanks & Regards
Pradeep


#3

You can cap ncpus or cores to a group as below, but not the compute nodes:

For example:

Limit the total number of ncpus that can be consumed by jobs in the “radioss” queue
to 100 per group; limit group1 to 50 ncpus, group2 to 25 ncpus, and set the overall limit
to 200 ncpus

qmgr –c "s q radioss max_run_res.ncpus ='[g:PBS_GENERIC=100],[g:group1=50],[g:group2=25],[o:PBS_ALL=200]'"

Other work around:

  1. Create a queue , say for example groupA
  2. assign acl_group to the queue for that queue
  3. Create a custom resource with the group name “grouping” as the host level resource
    • qmgr -c “create resource grouping type=string_array,flag=h”
    • add grouping to the resources: " … , grouping" of the $PBS_HOME/sched_priv/sched_config
    • kill -HUP ; so that updated configuration is read
  4. qmgr -c “set queue groupA default_chunk.grouping=groupA”
  5. for i in list of 7 nodes ; do qmgr -c “set node $i resources_available.grouping=groupA”
  6. So only that group of users will be able to submit jobs to that queue (groupA) and they will only be running only on those seven nodes.

Hope this helps


#4

Since you want it for User Group, you can use, “max_group_res”

Something similar to,

qmgr -c ‘set server max_group_res.ncpus=48’

Or, if you want Q specific

qmgr -c ‘set queue newq max_group_res.ncpus=36’


#5

Thanks Adarsh for support


#6

Thanks, max_group_res.ncpus not working in 18 version, its old and deprecated