Yes, Fermilab has done this as part of the HEPCloud project. See Burt Holzman's talk at last year's Condor Week for details. http://research.cs.wisc.edu/htcondor//HTCondorWeek2016/presentations/ThuHolzman_HepCloud.pdf There are others who have done it as well. We have the machines call back to the pool at Fermilab. There are other people who have set up a separate pool and flock to it. See my talk at Condor Week this year for details on how we did the same in Google Compute Engine.
From: HTCondor-users <htcondor-users-bounces@xxxxxxxxxxx> on behalf of Taha Vasi <tvasi@xxxxxxxxxxxxx>
Sent: Friday, March 10, 2017 10:16:42 AM
Subject: [HTCondor-users] condor and AWS
Has anyone implemented a condor pool, with bursting into AWS to eliminate peaks of jobs?
So the idea is to spin up machines in AWS when the job Q internally gets above a certain amount.. till the Q drops below some threshold.
Would it make sense to setup a separate condor pool in the cloud and then flock to it? Or just spin up AWS instances and have them join the existing pool whenever you determine?