VM Migration that Assures and Maximizes Accuracy in Multi-stage Information Processing Systems

In this paper, we propose a Virtual Machine (VM) migration method that maximizes the total accuracy in a multistage information processing system, under the constraint that the given ratio of tasks satisfy the constraint of improving the given level of accuracy (i.e., progress in information processing) within a required response time. Conventional methods only aim to improve the total accuracy and do not consider achieving the given level of accuracy for each task. The proposed method aims to improve the ratio of tasks that can satisfy the given level of accuracy by balancing the load among edge servers as much as possible. In order to confirm the effectiveness of the proposed method, we set the given level of task accuracy and evaluate the total accuracy by varying the number of VMs present in the system. The evaluation results show that the proposed method improves the number of VMs satisfying the given level of accuracy by 120% and the total accuracy by up to approximately 13% compared to the case without VM migration.