txcom2003@xxxxxxxxxxxxxxxxxxx wrote:
Hi, I have a question about MPI job. When I submitted MPI job, the change between Idle state to Running State took long time (about 5 to 10 minutes). Condor seems need some time to get the dedicated resource being claimed. But other condition happen, when I issued command condor_hold and condor_release, a few second after that, job state can change quickly. Does anyone know why this happen ?
The negotiation cycle for Condor happens periodically, default is every 5 minutes.
http://www.cs.wisc.edu/condor/manual/v7.1/3_3Configuration.html#10310It's possible that it took more than one negotiation cycle for the Schedd to get enough claims to start your job.
If your question is why did your job start up quickly the second time, it is possible the Schedd still had those claims to start your job up quickly after the hold+release.
If your question is why does your job go on Hold or become Idle after being release quickly, it's because the Schedd can make those changes directly without having to rely on other parts of the system.
Best, matt