Re: [HTCondor-users] Getting HTCondor to distribute jobs more "evenly" to machines in the pool.

[John M Knoeller <johnkn@xxxxxxxxxxx>]

If you configure

 

CLAIM_PARTITIONABLE_LEFTOVERS = false

 

Then only 1 job will be matched to each machine per negotiation cycle.  This will effectively spread the

jobs across the machines if the machines are all idle or partially idle.

 

But you should be aware that this is not a wise policy for a busy pool. For most HTCondor pools, the supply

of jobs vastly exceeds the number of machines to run them on, so all of the machines will be filled with

jobs nearly all of the time. 

 

CLAIM_PARTITIONABLE_LEFTOVERS = true

 

Gets your jobs started faster, and on a busy pool, that is usually matters more.

 

-tj

 

From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> On Behalf Of Andrew Cunningham
Sent: Wednesday, August 7, 2019 6:13 PM
To: Condor-Users Mail List <condor-users@xxxxxxxxxxx>
Subject: [HTCondor-users] Getting HTCondor to distribute jobs more "evenly" to machines in the pool.

 

Hi,

I notice that when submitting , say 8 jobs into a pool where >100 slots are available on a grid of multi-core Windows workstations, then HTCondor seems to typically assign all 8 jobs to a single multi-core machine ( which has a total of 16 slots) rather than "spreading" jobs around to various machines in the pool( all machines are pretty much the same).  Is there some way to configure HTCondor to distribute jobs more "randomly" around the available hardware.

 

Thanks

Andrew