[Condor-users] Relative performance with and without significant attribute clustering for specific situations?

["Ian Chesal" <ICHESAL@xxxxxxxxxx>]

Has anyone on the Condor team ever measured the benefits of significant
attribute clustering where you have a lot of 1 process clusters in the
system? Lets say every job has the exact same requirements, what's the
overhead in scheduling if I have:

a) One cluster with 1000 processes; versus
b) 1000 clusters with 1 process each, with auto-clustering turned on;
versus
c) 1000 clusters with 1 process each, with auto-clustering turned on and
a very specific set of significant attributes defined; versus
d) 1000 clusters with 1 process each, but auto-clustering turned off;

Obviously (d) should be the worst case and (a) should be the best case
but where to (b) and (c) sit between those two extremes? Even if you've
just got a feeling for the relative performance between those four
scenarios I'd appreciate knowing about it.

Thanks!

- Ian

--
Ian R. Chesal <ichesal@xxxxxxxxxx>
Senior Software Engineer

Altera Corporation
Toronto Technology Center
Tel: (416) 926-8300