Maintenance Scheduling for Cloud Infrastructure with Timing Constraints of Live Migration

Abstract

In this paper, we propose an implementation of a maintenance scheduler for cloud infrastructures. Live migration associated with maintenance work is important to ensure service continuity for all virtual machines in an infrastructure. However, executing the migration process when the machines are under heavy load negatively affects cloud users' businesses, such as by degrading performance and extending downtime. We can avoid this by finding an appropriate time period for live migration and performing the migration then. This idea is convenient for cloud users but inconvenient for cloud providers, that is, maintenance work should be completed as soon as possible for security reasons. To satisfy both the users' convenience and providers' requirements, we designed a maintenance scheduling problem to find the appropriate time period and to shorten the maintenance work period. Since it is a large-scale combinatorial optimization problem with complex constraints on maintenance requirements, we described the constraints by using answer set programming and implemented a maintenance scheduler on the basis of a divide-and-conquer approach to reduce the computational complexity exponentially. We evaluated our scheduler by using information on a real configuration of a commercial cloud infrastructure. While a naive approach to solving the maintenance scheduling problem could not find any feasible solutions within a realistic amount of time and memory, our implementation generated the best maintenance schedule for 1032 physical machines and 14208 virtual machines in 206 s with a memory usage of 1086 MB.