Adaptive large neighborhood search heuristics for multi-tier service deployment problems in clouds

Abstract

This paper proposes adaptive large neighborhood search (ALNS) heuristics for two service deployment problems in a cloud computing context. The problems under study consider the deployment problem of a provider of software-as-a-service applications, and include decisions related to the replication and placement of the provided services. A novel feature of the proposed algorithms is a local search layer on top of the destroy and repair operators. In addition, we use a mixed integer programming-based repair operator in conjunction with other faster heuristic operators. Because of the different time consumption of the repair operators, we need to account for the time usage in the scoring mechanism of the adaptive operator selection. The computational study investigates the benefits of implementing a local search operator on top of the standard ALNS framework. Moreover, we also compare the proposed algorithms with a branch and price (B&P) approach previously developed for the same problems. The results of our experiments show that the benefits of the local search operators increase with the problem size. We also observe that the ALNS with the local search operators outperforms the B&P on larger problems, but it is also comparable with the B&P on smaller problems with a short run time.