The Elasticity and Plasticity in Semi-Containerized Co-locating Cloud Workload

Abstract

Cloud computing with large-scale datacenters provides great convenience and cost-efficiency for end users. However, the resource utilization of cloud datacenters is very low, which wastes a huge amount of infrastructure investment and energy to operate. To improve resource utilization, cloud providers usually co-locate workloads of different types on shared resources. However, resource sharing makes the quality of service (QoS) unguaranteed. In fact, improving resource utilization (IRU) and guaranteeing QoS at the same time in cloud has been a dilemma which we name an IRU-QoS curse. To tackle this issue, characterizing the workloads from real production cloud computing platforms is extremely important. In this work, we analyze a recently released 24-hour trace dataset from a production cluster in Alibaba. We reveal three key findings which are significantly different from those from the Google trace. First, each online service runs in a container while batch jobs run on physical servers. Further, they are concurrently managed by two different schedulers and co-located on same servers, which we call semi-containerized co-location. Second, batch instances largely use the spare resources that containers reserved but not used, which shows the elasticity feature of resource allocation of the Alibaba cluster. Moreover, through resource overprovisioning, overbooking, and overcommitment, the resource allocation of the Alibaba cluster achieves high elasticity. Third, as the high elasticity may hurt the performance of co-located online services, the Alibaba cluster sets bounds of resources used by batch tasks to guarantee the steady performance of both online services and batch tasks, which we call plasticity of resource allocation.