condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of Thomas Materna
Sent: Monday, September 19, 2005
Subject: [Condor-users] Flocking
I have major problems with flocking. I have a pool A
of 3 computers sharing a filesystem. I have a pool B of 20 computers not
sharing the same file system as the pool A. A flocks to B. I have a bunch of
jobs submitted from A in standard universe but I have a very bad priority since
I've been doing that a lot lately. Another user also has whole bunch of
jobs submitted to A. But his are in vanilla universe, he added in his submit
file a requirement of the type
((Machine==A1) || (Machine==A2)...) where A1, A2 are the
machines in the pool A.
Well, he will never run on pool B, but he prevents me from
running on it!!!! What happens is that at every cycle, having a better
priority, he claims all the machines in pool B, my jobs can hence not do so.
Only then the jobs reject the machines for not meeting the requirement. I have
20 machines doing nothing!
How can I get around that? Is there a way to avoid the jobs
claiming machine they won't accept to run on anyway? If not, I consider it a
major flaw. One user like that could stop everything. And since he's not
running, his priority won't go up. I'll have to wait 5 days to have my priority
back to a competitive level. And even then, some machines won't be used.
Thanks for any help you guys could provide,