Arvados

Want to be able to support fractional CPU requests.

AWS nodes like t2.small have harsh CPU quotas, if you go about 20% (?) CPU utilization you start using credits, and you get 12 minutes of 100% CPU usage per hour. This works out to enforced 20% utilization on average.

It would be useful to be able to say, this node provides 0.2 VCPUs, so that when we ask for 1 VCPU we don't get placed on a t2.small that is actually much less than 1 VCPU of compute power -- and also have the ability to request fractional compute for lightweight jobs.