Seeding-Based Multi-Objective Evolutionary Algorithms for Multi-Cloud Composite Applications Deployment

Abstract

There are an increasing number of enterprises deploying their application services to multi-cloud to benefit the advantages brought by cloud computing. The multi-cloud composite applications deployment problem (MCADP) aims to select proper cloud resources from multiple cloud providers at different locations to deploy applications with shared constituent services so as to optimize application performance and deployment cost. Multi-objective evolutionary algorithms (MOEAs) can be utilized to find a set of trade-off solutions for MCADP. During population initialization of MOEAs, seeding strategies can considerably improve the algorithms’ performance. For example, the seeding-based MOEAs, AO-Seed and SO-Seed, introduce a pre-optimization phase to search for solutions to be embedded into the initial population of MOEAs. With the extra optimization overhead, however, the two seeding-based MOEAs can only identify one or a limited few solutions to MCADP utilized by MOEAs. To solve MCADP effectively and efficiently, we propose new seeding-based MOEAs in this paper. The approach can construct application-specific seeds according to problem domain knowledge and build a group of diverse and high-quality solutions for the initial population of MOEAs. Extensive experiments have been conducted on a real-world dataset. The results demonstrate that the proposed seeding-based MOEAs outperform SO-Seed and AO-Seed with less computation cost for MCADP.