Quasare

Abstract

Cloud computing promises flexibility and high performance for users and high cost-efficiency for operators. Neverthe- less, most cloud facilities operate at very low utilization, hurting both cost effectiveness and future scalability. We present Quasar, a cluster management system that in- creases resource utilization while providing high application performance. Quasar employs three techniques. First, it does not rely on resource reservations, which lead to underutiliza- tion as users do not necessarily understandworkload dynam- ics and physical resource requirements of complex code- bases. Instead, users express performance constraints for each workload, letting Quasar determine the right amount of resources to meet these constraints at any point. Sec- ond, Quasar uses classification techniques to quickly and accurately determine the impact of the amount of resources (scale-out and scale-up), type of resources, and interference on performance for each workload and dataset. Third, it uses the classification results to jointly perform resource alloca- tion and assignment, quickly exploring the large space of options for an efficient way to pack workloads on available resources without violating QoS constraints. Quasar moni- tors workload performance and adjusts resource allocation and assignment when needed. We evaluate Quasar over a wide range of workload scenarios, including combinations of distributed analytics frameworks and low-latency, stateful services, both on a local cluster and a cluster of dedicated EC2 servers. Quasar improves resource utilization by 47% in the 200-server cluster, while meeting performance con- straints for workloads of all types.