Two-level Hierarchical Cluster-Node Scheduling for Heterogeneous Datacenters
TimeTuesday, July 12th6pm - 7pm PDT
LocationLevel 2 Lobby
DescriptionTo address the increased heterogeneity, we present Qin, a two-level hierarchy scheduler for heterogeneous datacenters. The proposed approach combines cluster and node level scheduling algorithms, which can consider specific optimization objectives including jobs completion time, energy usage, and energy delay product (EDP). The novelty of the proposed scheduling is that it directly considers interference and heterogeneity in the scheduling process. Experiments on a real cluster and simulations with both real-world and synthetic workloads demonstrate the proposed scheduling approach outperforms state-of-the-art schedulers. This is the first work that considers an integrated cluster and node levels scheduling approach for heterogeneous datacenters.