Fair scheduling of multiple resource types over multiple and heterogeneous resource pools ∗

We consider the problem of scheduling a group of heterogeneous, distributed processes to a group of heterogeneous resource pools (physical servers, clusters of processors/cores) each with different types of resources. The processes specify their preference-constraints [20] and linearly elastic resource needs for heterogeneous resource pools. Our setting is a neutral, public cloud datacenter [6]. We extend the results of [12] on proportional and Max-Min Fairness (MMF) to this problem setting. A compromise is sought between a kind of proportional fairness across all resource pools, which may lead to not fully booked resources, and weighted Dominant Resource Fairness (DRF) [5] within each pool. We also give example models of subadditive elastic demand.