Tide: achieving self-scaling in virtualized datacenter management middleware

The increasing popularity of system virtualization in datacenters introduces the need for self-scaling of the management layer to cope with the increasing demands of the management workload. This paper studies the problem of self-scaling in datacenter management middleware, allowing the management capacity to scale with the management workload. We argue that self-scaling must be fast during workload bursts to avoid task completion delays, and self-scaling must minimize resource usage to avoid resource contention with applications. To address these two challenges, we propose the design of Tide, a self-scaling framework for virtualized datacenter management. A salient feature of Tide is its fast capacity-provisioning algorithm that supplies just-enough capacity for the management middleware. We evaluate the effectiveness of Tide with both synthetic and real world workloads. Our results show that the self-scaling capability in Tide can substantially improve the throughput of management tasks during management workload bursts while consuming a reasonable amount of resources.