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

Re: [Condor-users] GCB Performance



>Let me add more questions to what Junaid asked:
>
>1. If things happened normally, there must have been 3 rounds of
>"matching and execution" with 20 VMs but 2 rounds with 40 VMs. Is this
>what you saw? If you have the same or similar length of critical path in
>the pipeline for some reason such as match failure and serialized
>database access, you will get the same completion time.
>2. Does each job run for roughly the same period both with 20 and 40 VMs?
>
>You may also want to give us user logs for both cases (20VMs and 40VMs).
>
>Thank you.
 
I seen all the VMs machines match. /hosts/*/log/StarterLog.vm* indicated that each
machine is working away as the log files are incremented with data (GCB debug etc).
 
I am running 50 jobs - shell script which then runs an executable 100 times. the executable
always take the exact same time to run as its doing the exact same work but just with
different input data.
 
Is it possible that each condor job (which is a batch of 100 sub jobs) is too short in
its processing time that its negative on performance. Maybe I should try 10 jobs of 500
sub jobs and see if performance is improved?
 
Chris