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

Re: [HTCondor-users] Why all jobs are sent to slots in the same machine?



On 05/15/2013 01:00 PM, David Hentchel wrote:
...  I only ever want to allow 1 job running on a
> machine, but since these are 8 core systems, I currently see 8 Slots for
> each host.

Try this:
https://htcondor-wiki.cs.wisc.edu/index.cgi/wiki?p=WholeMachineSlots

In my case, I'd like the default matching algorithm to go 'hosts first'
rather than 'slots first' as it seems to do now.

The rationale is that 8 jobs running on 8 hosts each get all of host's
ram and i/o. Whereas 8 jobs running on 8 cores on the same host get
contention. Obviously, if your other 7 hosts have cpus from last century
or if your 8 jobs are pure cpu-bound & all they ever need fits
completely in the registers, you'd use the 'slots first' default.

-- 
Dimitri Maziuk
Programmer/sysadmin
BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu

Attachment: signature.asc
Description: OpenPGP digital signature