Learning-based Variable Neighborhood Search Algorithm for Cloud Service Deployment Problem with Time Windows

Fanchao Meng, Yang Lu, Weigong Lv, Dianhui Chu

Abstract: When services are deployed in a cloud environment, the selection of virtual machines mostly relies on the historical experience of service providers, which leads to a high cost of cloud resource. Additionally, such problem rarely considers the impact of service time on the cost of virtual machines. In this paper, we introduce and model the cloud service deployment problem with time windows (CSDPTW), wherein the cost of virtual machines is influenced by the usage time windows. To solve this problem, we propose a learning-based variable neighborhood search (L-VNS) algorithm. The algorithm first employs a hierarchical clustering algorithm to generate the initial solution. Then, a Q-learning-based reinforcement learning algorithm is used to optimize the selection of operators, which includes two neighborhood operators and three local search operators. To evaluate the performance of our proposed L-VNS, genetic algorithm (GA), variable neighborhood search (VNS), iterated local search (ILS) and the commercial solver CPLEX are applied to CSDPTW. Computational experiments on 20 test instances demonstrate that the proposed L-VNS algorithm can find better solutions than GA, VNS, ILS and CPLEX.