Re: [Condor-users] spreading I/O bound processes across a cluster

["Little, Colin E" <ColinLittle@xxxxxxxxxxxxx>]

Armen,

Perhaps others will have a better solution, but my quick and dirty fix
to a similar problem was to have the jobs rank machines by
VirtualMachineID.  That way they prefer to spread themselves out rather
than clump together.  It's not perfect by any means, but it seems to
work fairly well.

-Colin Little
ColinLittle@xxxxxxxxxxxxx




> -----Original Message-----
> From: condor-users-bounces@xxxxxxxxxxx [mailto:condor-users-
> bounces@xxxxxxxxxxx] On Behalf Of Armen Babikyan
> Sent: Wednesday, April 19, 2006 10:57 AM
> To: Condor-Users Mail List
> Subject: [Condor-users] spreading I/O bound processes across a cluster
> 
> Hi,
> 
> Suppose I have 12 quad-processor machines, each with a large local
disk
> and a commonly-mounted NFS partition.  I setup a big Condor
experiment,
> where, 20 processes will (at some stage, most often simultaneously) be
> doing a lot of I/O to the aforementioned large local disk.
> 
> Is there a simple way to get Condor to evenly spread this I/O bound
> process over all 12 of the machines?  Condor sees a 48-processor
system,
> and sometimes schedules 4 of these processes on one machine, and none
on
> another.  I'm not even considering the possibility of exhausting local
> diskspace yet - I just want a more-or-less even distribution of I/O
over
> my cluster's component I/O subsystems.  Any ideas? :-)
> 
> Thanks!
> 
>   - Armen
> 
> --
> Armen Babikyan
> MIT Lincoln Laboratory
> armenb@xxxxxxxxxx . 781-981-1796
> 
> _______________________________________________
> Condor-users mailing list
> Condor-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
This email and any files transmitted with it are confidential, proprietary
and intended solely for the individual or entity to whom they are addressed.
If you have received this email in error please delete it immediately.