Improving VM Scheduling Policies in Cloud Computing Through Quantum-Assisted Reinforcement Learning

Cloud computing refers to executing tasks submitted by customers using Virtual Machine Scheduling Policies (VMSPs). Currently, these VMSPs of the cloud lack intelligence mechanisms, and it becomes necessary to impart modern intelligence techniques to enhance them and ultimately improve cloud performance. This paper evaluates the performance of these VMSPs by conducting experimental-based research considering VMSPs including Time Shared (TS_VMSP), Space Shared (SS_VMSP), and Dynamic Workload Shared (DWS_VMSP). Tasks are processed and executed in multiple scenarios using these VMSPs on Xen and Kernel-based VM (KVM) hypervisors. The performance parameters considered for the evaluation of these VMSPs include Average Start Time (Avg_S_T), Average Finish Time (Avg_ F_T), Average Turn Around Time (Avg_ T_ A _ T) and Average Success Percentage (Avg_Success _ %). The consistency of these VMSPs is also evaluated from the experimental results using the Linear Regression Model across all the performance parameters. The experimental results convey that the TS_VMSP outputs ideal performance in terms of Avg_S_T, Avg_F_T, and Avg_T_A_T, and SS_VMSP outputs ideal performance in terms of Avg_Success_% for both the hypervisors. The consistency evaluation results output SS_VMSP to provide ideal performance in terms of Avg_S_T, Avg_F_T, and Avg_T_A_T for both hypervisors. Additionally, the SS_VMSP and DWS_VMSP provide ideal performance concerning consistency evaluation for the Avg_Success_ % performance parameter. Recognizing that no single VMSP outperforms others, the authors propose a novel architecture that uses a hybrid VMSP approach by integrating TS_VMSP, SS_VMSP, and DWS_VMSP with VMs having Quantum-assisted Reinforcement Learning model. This architecture provides an intelligent mechanism and accelerates task computing, ultimately improving overall cloud performance.