[HTCondor-users] Dynamic slot/machine job filling behavior

[Bob Ball <ball@xxxxxxxxx>]

Our high-core-count machines are running Dynamic cpu allocation. But, in 
the instance that no 8-core jobs are hanging around to run, our 
priorities swap to allow single core jobs to run.  And, here is the 
rub.  First machine A is filled to a great extent, then B, then C, where 
what I would really like done is to pass the first job to A, the next to 
B, the next to C, and so on.  But, I cannot find a way to do that?  In 
the past I could just use SlotId to set up the Rank expression, but that 
is always 1 in the context of the negotiator, so they seem to allocate 
away happily until they are full, or nearly so, then move on to the 
next, and so on.  This is a really bad way to start a bunch of jobs at 
once.

What I'm really looking for is a reversal of the negotiator logic 
revision described here:
https://htcondor-wiki.cs.wisc.edu/index.cgi/tktview?tn=3435

Has anyone encountered this issue?  I understand FNAL is now 
understanding that this could be an issue.  Does anyone have a solution?

Thanks much,
Bob Ball