[torqueusers] big problems with CPU allocation

bill cluster.bill at alinto.com
Thu Sep 21 03:53:51 MDT 2006


my cluster does weird things.

1. It does not allocate enough CPU:
qstat -f on job show me:
    exec_host = 
So its 11 CPU, but:
    Resource_List.nodes = 3:ppn=4

This is reproductible. I asked for 8 CPU:
     exec_host = eclssgh5/3+eclssgh5/2+eclssgh5/1+eclssgh5/0
     Resource_List.nodes = 2:ppn=4

Why torque gives me not the good amount of CPU?

2. Strangely, I switch user, and with another uid, it works (ie, it 
gives the good amount of CPU)

3. torque gives more job to nodes than allowed:
      state = free
      np = 4
      ntype = cluster
      jobs = 0/253.mclssgh1, 0/242.mclssgh1, 1/242.mclssgh1, 
1/234.mclssgh1, 2/243.mclssgh1

4 CPU, but 5 jobs?

Users on this cluster request only nodes and CPU with
#PBS -l nodes=x:ppn=y

What is causing this bug?

Thanks for any help

More information about the torqueusers mailing list