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

Re: [Condor-users] condor 6.7.6 load limit?

On 9/1/05, Jeffrey McDonald <jeffrey.mcdonald@xxxxxxxxxxx> wrote:
> Hello,
> I've setup my local cluster machines with condor.   I haven't done very
> much configuration, only setup a central machine which negotiates and
> collects the requests for jobs and job submission is possible from any
> member node.  Everything works with this, jobs can be submitted, they
> are run and are successful.
> However, one problem that remains is that none of the condor submitted
> jobs generate a load of more than 30% on the CPU that it is running on.
> This is regardless of whether the machine has interactive activity or
> not.    The net result is that jobs that take 20 minutes to run at a
> high priority take 4-5 hours.

condor has nothing to do with this (assuming your machines have no
other significant tasks running on them in which case forcing the
processes to a higher priority could help)

if you run a single job by hand on one of those machines how much CPU
load is there?

if you run several jobs on different machines by hand how much usage is there?

if there is about 30% in the first instance your jobs are not CPU
limited on the execute nodes.

if the 30% load only occurs when multiple jobs are running then the
jobs as they currently stand are not benefitting much from being
parallelized due to contention for some shared resource. this is
normally IO related and is not something condor can 'fix' for you...