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

[Condor-users] A question about Backfill Jobs

Can I do backfill with a non-BOINC program? Here's my scenario:

We use a good portion of our farm for runtime analysis of our software
and it's algorithms. Generally we use user+kernal measurements but
pre-release we need to do "user experience" measurements that check wall
clock run times (it helps take in to account things like shared library
load and unload times and so on).

We do these measurements over our pool running 1 instance of our
software per core/cpu. Problem is: as an experiment tapers off you end
up with fewer jobs to run and machines get less and loaded. So the tail
end of an experiment is skewed towards more favourable run times. We do
a lot of reptitions of the measurement to bring the noise down but we
could cut the repetitions down if we could ensure the machines were
always loaded with similar work to our jobs, even at the tail end of the

I was thinking the backfill feature might be a nice way to do this. Some
things that aren't clear from the manual section: does Condor spawn one
backfill process per-slot on the machine? So as each slot in my 4-slot
system becomes free I get another backfill job running? Or is 1 backfill
job per-machine?

Also, *must* I run BOINC? Is there something specific about the BOINC
client that the BOINC backfill methods require to operate properly?
Could I make my custom backfill loading job look like a BOINC client and
fake it out?

- Ian

Confidentiality Notice.  This message may contain information that is confidential or otherwise protected from disclosure.
If you are not the intended recipient, you are hereby notified that any use, disclosure, dissemination, distribution, 
or copying of this message, or any attachments, is strictly prohibited.  If you have received this message in error, 
please advise the sender by reply e-mail, and delete the message and any attachments.  Thank you.