A Dynamic Resource Manager with Effective Resource Isolation Based on Workload Types in Virtualized Cloud Computing Environments

Abstract

To use computing resources for processing parallel algorithms on demand, cloud computing has been widely used since it is able to scale in response to load increases and decreases. Typically, cloud computing providers offer virtual machines to cloud users with static configurations, and these configurations are not changed until virtual machines are shutting down. To accelerate parallel processing computations in cloud computing environments, we design and implement a dynamic resource manager by isolating resources based on workload types. To avoid unnecessary context switching and increase CPUs affinity, our dynamic resource manager determines whether vCPU to physical CPU core pinning is required. If so, the VM’s vCPUs are pinned by our dynamic resource manager, which can guarantee the resource and performance isolation. With our proposed resource manager for virtual machines, we can achieve a performance boost and load balancing at the same time. Performance results show that our proposed method outperforms the default scheduler of Xen about 36.2% by reducing the number of context switching for VMs.