Virtual Machine Placement Strategy for Self-driving Using Cluster-based Genetic Algorithm

Abstract The problem of virtual machine (VM) live migration in self-driving systems targets reallocating resources among VMs running different self-driving services for load balance. It is of great importance to enable a running VM to be moved to another physical machine (host) seamlessly. Due to the nature of self-driving applications, it is important to select an optimal set of hosts to place VMs within an interactive time. The VM placement problem can be formalized as a bin packing problem, which is proved to be NP-hard. To solve the problem, we develop a cluster-based genetic algorithm that outputs an approximation result of the bin pack problem. In particular, our proposed algorithm clusters the population of current generation and selects individuals from different groups with reduced crossover operations. The number of crossover operations is directly related to the algorithm efficiency. We use the run-time features to evaluate the preference of VMs on hardware resources, which is utilized to generate initial solutions and avoid overload. Experimental results show that our approach is able to outperform the tradition genetic algorithm regarding both accuracy and efficiency.