[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[HTCondor-users] Desktop configuration: startd vs schedd.

Hi all,

Here, this is our context:
-We have 2 kinds of machines: cluster and workstation. Users are submitting jobs from their workstations. Due to our small number of cluster, we choosed to allows workstations to compute jobs.

Our jobs take a lot of memory and the workstation may swap all the time. We did implement a requirement for memory usage (given the parameter memory in job submission). But the parameter seems to check the machine memory and not the available memory. Is there a way to change that ?

We would like to amend our START expression, to forbid job's starting while the "current" user of the machine is submitting many jobs or when machine has jobs in its queue schedd).
Here, the chain of events that happens:

1/ from Workstation 1 , user 1  submits 400 jobs to ht condor
2/ ht condor starts (the first) 20 jobs from user1 on computers (cluster and workstation).
3/ workstation 1 gets jobs from htcondor (from user1 or someone else).
4/ workstation 1 starts to swap because the job require the whole memory of the machine. 5/ Workstation and all condor daemons stop working. So, jobs from user 1 can not be computed any more.

I tried something like this:
START = ( ( $(CPUIdle) || (State != "Unclaimed" && State != "Owner")) && JobsSubmitted==0 && TotalIdleJobs == 0 && TotalRunningJobs == 0)

but as I read in the documentation, those fields could not be used into the START expression