Hi,
I have a few queues and like to have nodes 'attached' to them with
faster nodes being able to take overflow from the slower queues, in the
past I have used something like:
# Tie slower quad phenoms to medium64 queue
SRCFG[medium64]
HOSTLIST=a01,a02,a03,a04,a05,a06,a07,a08,a09,a10,a11,a12,a13,a14,a15,a16,a17,a18,a19,a20,l01,l02,l03,l04,temp1
SRCFG[medium64] CLASSLIST=medium64
SRCFG[medium64] PERIOD=INFINITY
SRCFG[medium64] RESOURCES=PROCS:-1
# Tie the parallel queue to the faster quad core phenoms cpus, allow
medium64 overflow
SRCFG[parallel]
HOSTLIST=f01,f02,f03,f04,f05,f06,f07,f08,f09,f10,d01,d02,d03,d04,d05,d06,d07,d08,d09,d10,k01,k02,k03,k04,k05,k06,
k07,k08,k09,k10
SRCFG[parallel] CLASSLIST=parallel,medium64-
SRCFG[parallel] PERIOD=INFINITY
SRCFG[parallel] RESOURCES=PROCS:-1
However it now seems that the negative affinity of medium64- is not
honored, if we suibmit a job to the medium64 queue it always seems to go
to a f,d or k node even when there are many a nodes that are few.
Perhaps I am doing something wrong but is it possible that the negative
affinity no longer works and all the affinities are equal now and if so
can it be reverted back to the old way?
showconfig shows this:
SRCFG[medium64] CLASSLIST=medium64 HOSTLIST=a01 a02 a03 a04 a05 a06
a07 a08 a09 a10 a11 a12 a13 a14 a15 a16 a17 a18 a19 a20 l01 l02 l03 l04
temp1
SRCFG[parallel] CLASSLIST=parallel:==medium64 HOSTLIST=f01 f02 f03 f04
f05 f06 f07 f08 f09 f10 d01 d02 d03 d04 d
05 d06 d07 d08 d09 d10 k01 k02 k03 k04 k05 k06 k07 k08 k09 k10
Any help gratefully received,
Jon
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers
