Efficient Hybrid Central Processing Unit/Input-Output Resource Scheduling for Virtual Machines

The use of virtualization technology in industrial control is increasing. However, virtual instances or virtual machines (VMs) are confronted with the unreasonable resource allocation in the industrial control cyber range, thereby highlighting the increasing importance of resource scheduling. In the Xen open source system, the input-output (IO)-intensive task response is extended because the system does not distinguish between CPU-and IO-intensive tasks. Therefore, this article presents an improved task performance through the hybrid scheduling of CPU and network IO resources. This method uses Cap-and Timeslice-scheduling algorithms for CPU resource scheduling. First, the Cap-scheduling algorithm uses historical data to train a recurrent neural network model for classification and then utilizes the heuristic method to set the upper limit of cap value for VMs. Second, the Timeslice-scheduling algorithm adjusts the timeslice based on Q-learning to shorten the execution time of the overall tasks. This article proposes an IOb-scheduling algorithm for network bandwidth scheduling. The part that does not exceed the average bandwidth is eliminated and distributed to other VMs by monitoring the bandwidth usage of each VM, thereby improving the utilization of bandwidth. Experimental results showed that the proposed CPU/IO scheduling algorithms improved the overall benchmark performance substantially.