Evaluation and Optimization of a Session-based Middleware for Data Management
The current massive daily production of data has created a non-precedent opportunity for information extraction in many domains. However, this huge rise in the quantities of generated data that needs to be processed, stored, and timely delivered, has created several new challenges. In an effort to attack these challenges proposed a middleware with the concept of a Session capable of dynamically aggregating, processing and disseminating large amounts of data to groups of clients, depending on their interests. However, this middleware is deployed on a commercial cloud with limited processing support in order to reduce its costs. Moreover, it does not explore the scalability and elasticity capabilities provided by the cloud infrastructure, which presents a problem even if the associated costs may not be a concern. This thesis proposes to improve the middleware’s performance and to add to it the capability of scaling when inside a cloud by requesting or dismissing additional instances. Additionally, this thesis also addresses the scalability and cost problems by exploring alternative deployment scenarios for the middleware, that consider free infrastructure providers and open-source cloud management providers. To achieve this, an extensive evaluation of the middleware’s architecture is performed using a profiling tool and several test applications. This information is then used to propose a set of solutions for the performance and scalability problems, and then a subset of these is implemented and tested again to evaluate the gained benefits.