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

Re: [Condor-users] job suspension?



Michael Di Domenico wrote:
> I'm a little confused over the implementation of job suspension in
> condor 7.2, hopefully someone can clear it up for me.
> 
> We're running condor across RHEL5 workstations and a few clusters
> RHEL4/5 32/64bit, ia64, and Solaris at our company.
> 
> We want two behaviors to happen.
> 
> 1.  When a workstation user touches his keyboard, if there is a condor
> job it's suspended.
> 2.  If a user logs into a cluster and kicks off a job (via batch
> scheduler), the condor job is suspended on the nodes until the job is
> finished
> 
> So here's the questions:
> 
> 1.  In Condor v7.2 is condor_kbdd required on the workstations to
> detect activity on a linux workstation w/ USB key/mou?
> 2.  If condor_kbdd is not required, who is detecting this activity?
> 3.  If it's not condor_kbdd, under what circumstances does it consider
> a node (srv/wks) to be occupied? (ie what does it monitor)
> 
> We've considered creating a pre/post amble in our batch scheduler to
> alert condor to a pending job, but we're not sure if there might be a
> better way.
> 
> Thanks
> - Michael

1) it depends, often yes. If you can find a device that gets updated
when the user does something you can avoid the kbdd though.

http://www.cs.wisc.edu/condor/manual/v7.3/3_3Configuration.html#17529

2) the startd does detection.

3) have a look at:

http://www.cs.wisc.edu/condor/manual/v7.3/3_5Policy_Configuration.html

You can also use load average on the system, such as when your batch
system is running the CPU.

Best,


matt