Re: [HTCondor-users] One job per machine

[Greg Thain <gthain@xxxxxxxxxxx>]

On 11/10/22 11:38, Michael Thomas wrote:
> I've got a workflow that is very network intensive (verifying file 
> checksums from a distributed ceph filesystem). Two jobs running on 
> the same machine in the cluster can easily saturate a cluster node's 
> 1Gbps network connection. 


Hi Mike:


We can do this with HTCondor Custom Machine Resources. In the startd, 
add a knob like


MACHINE_RESOURCE_HeavyNetwork = 1


This means that this machine can only run one job where

Request_HeavyNetwork = 1


is in the submit file.


-greg