An Optimization Approach For SaaS Software Deployment Toward Performance Improvement

The environment SaaS software running is open and dynamic.It is possible that its performance will degrade in the process of long-time running.An effective solution to the problem is to optimize its deployment according to the changing environment.However,compared with the traditional software,the deployment of SaaS software is characterized by servitization in software virtualization in hardware and clustering in deployment.Therefore,the traditional deployment optimization algorithm cannot be used to solve the deployment optimization problem for SaaS software.This dissertation proposes a deployment optimization method considered the characteristics of the deployment of SaaS software,which can automatically optimize the deployment architecture of SaaS software at runtime,the details are as follows:(1)This dissertation proposes a SaaS software deployment description language(SSDL),which supports the performance optimization of SaaS software.SSDL is designed from five aspects,i.e.SaaS software,cloud platform,deployment plan,deployment constraints,and usage model,in order to support expressing the information on the deployment architecture of SaaS software and its optimization.SaaS software model describes the hierarchical combination of SaaS software by defining services and service connectors,and describes the dynamic binding,adaptation,selection and failure processing of service interfaces by defining complex behaviors for service connectors.Cloud platform model describes the resource types it provided for deploying SaaS software by defining virtual machine resources and virtual network link resources.Deployment plan model describes the allocations between the service instances of SaaS software and the virtual machine instances of cloud platform by defining deployment nodes and associating them with the services in SaaS model and virtual machine type in cloud platform model.Deployment constraint model describes the constraints that SaaS software needs to abide by defining relative position and absolute position constraints.Usage model describes the usage situation of SaaS software by defining workload and scenario behavior.(2)This dissertation proposes a method for transforming SSDL to queuing network model,and evaluating the deployment performance of SaaS software based on queuing network model.Firstly,according to the SSDL description information,the SaaS software,cloud platform,deployment plan,deployment constraints,and usage scenario is defined in formal method.Then based on these information,we construct a queuing network model with two-tier structure containing software layer and hardware layer,which are respectively used to model software execute process and deployment plan that influence the deployment performance of SaaS software.Finally,the solving formulas for deployment performance of SaaS software are given by solving the constructed queuing network model,and based on these formulas,the impact of different deployment changes on SaaS software are evaluated,and several deployment changing measures for optimizing the performance of SaaS software are given.(3)This dissertation proposes a method for optimizing the deployment plan of SaaS software based on genetic algorithm is proposed.First of all,based on the previous evaluation method for the deployment performance of SaaS software,aimed to minimize the end-to-end response time,maximum utilization and costs,a deployment optimization model of SaaS software is constructed.Combining the genetic algorithm and the previous conclusion resulted from the evaluation of the impact of deployment changes,an optimization algorithm to solve the deployment optimization model is given.For this algorithm,its basic operations are designed by adopting group-based coding way according to the deployment features of SaaS software.Based on the conclusion resulted from the evaluation of the impact of deployment changes,6 deployment optimization strategies are designed,which are integrated into traditional genetic algorithm as a new evolutionary operator.