Task-aware based co-scheduling for virtual machine system

Today, virtualization technique is increasingly mature and prevalent in server consolidation and HPC. Virtual machine monitor plays a significant role in the resource management by dynamically mapping the virtual CPUs of virtual machines to physical CPUs according to chosen scheduling policy. However, since a virtual machine monitor lacks the insight into each virtual machine, the unpredictability of each workload makes effective resource allocation difficult. Particularly, current virtual machine scheduling policy has a critical impact on the performance of concurrent workload due to the non-synchronization of virtual CPUs. This paper presents a task-aware co-scheduling scheduler for virtual machine system. The task-aware mechanism is based on inference techniques using gray-box knowledge which can infer the concurrency and synchronization of guest OS level tasks. With this inference, proposed scheduler schedules the designated virtual machine to make it possible that corresponding virtual CPUs in this virtual machine can run on the physical CPUs synchronously in order to reduce the cost of synchronization between processes or threads. All implementation is confined to the virtualization layer based on Xen virtual machine monitor and the Credit scheduler. We evaluated our prototype in terms of synchronization performance and CPU fairness over synthetic mixed workloads and realistic applications. The experiment results indicate that task-aware based co-scheduling policy is feasible to improve the performance of virtual machine system for concurrent tasks.