Department of Computer Engineering
MS THESIS PRESENTATION
Application Framework for Supporting Performance Isolation in Software as a Service Systems
Orhun Alp Oral
(Supervisor: Asst. Prof. Dr. Bedir Tekinerdoğan)
Computer Engineering Department
Scalability of a cloud application is the ability of an application that handles the service level agreement (SLA) requirements for all customers. In a non-isolated SaaS system, the different clients of a SaaS can freely use the resources of the SaaS. Hereby, disruptive tenants who exceed their limits can easily cause degradation of performance of the provided services for other tenants. To ensure performance demands of the multiple tenants usually elasticity of the SaaS is needed, which supports changing the hardware resources. In case the users demand change, the system can scale itself up or down according to the minimum necessary computational power to handle all the requests within the minimum requirements of SLA. The main goal of elasticity, however, is not directly to support performance isolation but scalability of the SaaS in general. As such, relying only on elasticity does not explicitly solve the problem of performance isolation. Performance isolated systems aim to ensure the minimum requirements of the SLA to all of the customers. In order to build a performance isolated system, it is necessary to ensure fair behavior of the system with respect to its customers. Various performance isolation approaches have been introduced including artificial delay, round robin, blacklist, and thread pool. However, these approaches assume a SaaS system with thin clients whereby all the tasks are handled by the SaaS.
This thesis introduces the Tork Framework, for developing scalable and performance isolated enterprise SaaS applications. Tork Framework provides a new way for developing a cloud application from scratch or converts existing software applications to scalable cloud applications. Tork Framework introduces built-in features applicable for the application layer, database layer and client layer. These features include techniques for security controls, logging, and multi-tenancy on relational database management systems (RDMS). Built-in server components of Tork Framework provide service oriented architecture for all business objects in the RDMS, and client components make the application work on different platforms. Also, this framework provides a cloud based IDE for configuring parameters and writing custom codes without the need of any other individual tool via online graphical user interface (GUI).
The framework as such provides an approach for supporting the design and realization of performance isolated cloud system by also considering the usage of resources of thick clients. To illustrate the framework and validate our approach we have adopted a real SaaS environment in which a case study is implemented using the Tork framework. Different Implemented performance isolation techniques are tested in this case study using both thin and thick client types.
DATE: 08 January, 2015, Thursday @ 09:00
PLACE: EA-516 (Bilvea)