【作者】 郝晓宁；

【导师】 刘士军；

【作者基本信息】 山东大学 ， 软件工程（专业学位）， 2014， 硕士

【摘要】 SaaS (Software as a Service)支持中小企业基于互联网的多租户行为,现在被广泛地用来进行软件服务的发布。在开发SaaS应用的过程中,为了满足租户的要求,同时尽可能地降低服务提供商的成本,需要考虑的重要一点就是应用的优化部署。从客户的角度来讲,SaaS应用必须要满足其对服务性能及其他指标的要求(例如分配的硬件资源是否足够),使所有用户满意。而站在服务提供商的角度来讲,需要尽可能的解决资源的过度分配或利用率过低等问题。这涉及到对系统可扩展性、未分配资源的利用以及升级成本这些因素的权衡。因此,多租户应用的优化部署成为改善服务性能以及提高资源利用率的关键。基准测试(Benchmark)是指通过设计科学的测试方法、测试工具和测试系统,实现对一类测试对象的某项性能指标进行定量的和可对比的测试,进而为应用系统的运行以及优化部署提供支持。供应商关系管理系统(Supplier Relationship Management)属于电子商务类的多租户SaaS应用,其主要用来改善供应链上下游的业务关系,整合制造商和供应商的资源优势,使制造商和供应商之间建立更加紧密、持久的伙伴关系,最终达到共赢的目的。目前,在众多的能够用来评估电子商务类应用的基准测试中,它们的针对性各不相同,例如业务型WEB基准TPC-W,再如模仿Ebay.com的拍卖型基准RUBiS以及公告栏型基准RUBBoS。在它们所应用的领域,这些测试基准被广泛用来评估和比较计算机系统的性能。然而,传统的基准测试并不适用于具有不同特征的所有应用。因此,我们需要开发一套针对多租户SaaS应用的测试基准,用来专门测试和评估该类应用。多租户SaaS应用的Benchmark系统以供应商关系管理(SRM)应用软件为测试基础,通过模拟对SRM的业务请求来评估服务器的性能以及应用本身的表现力,并以大量的实验数据为基础,得出关于SRM软件本身的规律性结论,同时为SRM应用的优化部署提供科学合理的支持,能够很好的反应SaaS应用的复杂性。本文在对自主开发的SRM类SaaS应用进行需求分析的基础上,选定了TPC-W Benchmark作为我们基准测试的参考,开发一套针对SRM应用的基准测试系统。按照面向对象的软件工程的要求,对系统的开发进行了前期的需求分析及系统架构的设计,将整个系统按功能进行模块的划分,在明确了各个模块的功能之后,分模块对系统进行详细设计,并以此为基础编码实现系统的功能。最后,为了验证基准测试系统的可靠性及可用性,我们设计了几组实验来进行测试。通过大量的测试,我们发现并总结了很多有用的结论,从而验证了我们的基准测试系统是可以用来为SRM应用的优化部署提供支持的。更多还原

【Abstract】 For SMEs, SaaS (Software as a Service) supports Internet-based multi-tenant behaviors and now is widely used for the release of the software services. In order to meet the tenant’s requirements and to minimize the cost of the service providers, it is an important point to optimize the deployment of SaaS applications in the process of developing them. From the customer’s perspective, SaaS applications must meet the requirements of service performance and other indicators, such as the adequacy of hardware resource allocation, in order to satisfy all the users. While standing on the service provider’s perspective, they need to try their best to address over-allocation or low resource utilization issues. This involves weighing factors of scalability of the system, unallocated resource utilization and costs of upgrading. Therefore, optimizing the deployment of multi-tenant applications becomes the key to improve services performance and to increase resource utilization.Benchmark is defined by the design of scientific testing methods, test tools and test systems, to achieve a quantitative and comparable test for one performance of a type of test object thus providing support for the operation and optimization of the deployment of application systems. Supplier Relationship Management System is a multi-tenant SaaS applications like e-commerce, which is mainly used to improve business relationships in the supply chain downstream. It integrates manufacturers and suppliers so that they can establish closer and enduring partnership and ultimately achieve a win-win goal.Currently, for many Benchmarks which can be used to evaluate the performance of e-commerce applications, they are designed for different purposes. Such as TPC-W is designed for business-based Web Benchmark, RUBiS is for auction-based Benchmark that mimic Ebay.com and RUBBoS for bulletin-based Benchmark. These benchmarks are widely used for evaluation and comparison of the performance of computer systems in a certain field. While, traditional Benchmark is not applied for all applications which have different characteristics. Therefore, we need to develop a Benchmarks suite for multi-tenant SaaS applications for specialized testing and evaluation of such applications.Benchmark suite for Multi-tenant SaaS applications is based on Supplier Relationship Management (SRM) application software. It can evaluate the performance of the servers and application by simulating the service request to the SRM application and it can reflect the complexity of SaaS applications well.We develop a benchmark system for type SaaS applications based on the demand analysis of the self-development SRM applications and we selected TPC-W Benchmark as our Benchmark reference. Accordance with the requirements of object-oriented software engineering, we conducted a preliminary requirements analysis and system architecture design. The whole system is divided into several function modules and each module is designed in detail. Finally, in order to verify the reliability and availability of the benchmark system, we designed several sets of experiments. Through extensive testing, we found a lot of useful conclusions, which illustrates that our benchmark system can be used to provide support for optimizing the deployment of SRM application.更多还原