云环境中面向服务软件的演化部署优化方法

Evolutionary deployment optimization for service-oriented software in cloud

针对现有的部署优化方法在求解云环境中面向服务软件的部署优化问题时,无法处理服务实例和虚拟机实例的伸缩以及无法保障求解质量等问题,本文提出了一种新的部署优化方法.该方法以提高面向服务软件的运行性能和降低运行成本为目标构建问题模型,并设计了一种基于遗传算法的MGA-DO算法对其进行求解.MGA-DO算法采用基于组的编码方式对软件的部署方案进行编码,然后结合基于组的单点交叉操作,实现了在优化过程中对服务实例和虚拟机实例的伸缩.此外,该算法引入现有的部署优化经验,设计了多种局部搜索规则,以进一步提高算法的求解性能.最后,一系列模拟实验表明,相比现有的算法,MGA-DO算法在求解所研究的问题时表现出了更好的性能.

This paper proposes a new deployment optimization method, to address the drawbacks of existing deployment optimization methods, for optimizing the deployment architectures of service-oriented software in cloud. Examples of these drawbacks are the lack of scalability of service instances and virtual machine instances, and the inability to guarantee the solving quality. The method first constructs a deployment optimization model with the goals of improving the running performance and reducing the operation cost of service-oriented software. Next, a genetic-based algorithm MGA-DO is utilized for solving the model. The MGA-DO adopts a group-based encoding scheme to encode the deployment architectures of service-oriented software and combines this scheme with a group-based crossover operator to realize the scalability of service instances and virtual machine instances in the optimization process. Moreover, the MGA-DO utilizes the existing knowledge of deployment optimization to design five types of local search rules, to further improve the local search ability of the algorithm and accelerate the convergence speed. Finally, a series of simulations show that, compared with the existing algorithms, the MGA-DO algorithm performs better on solving the research problem.