Re: [condor-users] New administrator questions

[Alan De Smet <adesmet@xxxxxxxxxxx>]

Mike Box <mbox-l@xxxxxxxxxxxxxxxxxxx> wrote:
> I recently created a Condor cluster of Unix computers whose
> primary purpose is to support an instructional classroom
> environment.  How do I configure Condor to favor the class
> times without adversely affecting the running job(s)?  I really
> want to avoid defining time-of-day windows to allow or prohibit
> processing.  I'm sure that there's got to be a more flexible
> approach.

I'm not clear on what you mean for Condor to favor class times.
Should Condor be more willing to use a machine while classes are
in session?  There are number of variables you could put into
place, perhaps in conjunction with the time-of-day windows.  More
extremely, you could use Condor's Hawkeye extensions to run a
shell script that will provide information used in the decision
making process (see below for more on this).

> Also, similar to the above situation, there is nightly
> administrative processing that causes the running job to be
> evicted from one computer to another.  Due to the schedule /
> timing of these administrative events, tests showed the job
> "bouncing" among computers with little or any real work being
> accomplished.  How can I configure Condor to recognize such
> processing and react more favorably?

Locally we have a macro called Maintenance which is true for a
particular window when automated scripts run early in the morning
(using ClockMin).  Various other macros make decisions based on
that.  One simple option would be to set PREEMPT to false if you
are in the maintenance window.  Jobs will be suspended during the
maintenance.

Here's a quick and dirty (and untested) example based on our
local configuration:

Maintenance  = (ClockMin > 255 && ClockMin < 315)

PREEMPT = (Activity == "Suspended") && $(ActivityTimer) > 300 \
          && $(Maintenance) == False

You could do more clever things like redefining CPUBusy to be
more tolerant of high loads during $(Maintenance).

On a more extreme level, the Hawkeye additions to Condor mean you
can run (nearly) arbitrary programs and include the results in
your machine classad.  You could use this to make policy
decisions based on anything you can describe in a shell script.
A shell script could grep for your nightly maintenance scripts
running and set the Maintenance flag based on that (although
there is a lag in that information be updated, so this is more
complex).

-- 
Alan De Smet                              Condor Project Research
adesmet@xxxxxxxxxxx                 http://www.condorproject.org/
Condor Support Information:
http://www.cs.wisc.edu/condor/condor-support/
To Unsubscribe, send mail to majordomo@xxxxxxxxxxx with
unsubscribe condor-users <your_email_address>