Efficient Deployment of Stream-based Distributed Systems on Cloud Infrastructure

Stream-based distributed systems are specialized data processing systems, where a continuous stream of data is pushed as input, such as by media sources, sensor networks or tracking services, that has to be processed and analyzed in a responsive way. Distribution often essentially means that system components are geographically distributed. This thesis was motivated by a stream-based, geographically distributed system, focusing on monitoring and analyzing broadcasted multimedia streams. Such systems have multiple unique characteristics that are currently not addressed by software architectures explicitly, such as geographical scalability or supporting continuous transfer of high load of data. This thesis offers two major contributions. From a research significance perspective this thesis provides a systematic review of state of the art approaches and methods in the domain of software architecture design and deployment models. Firstly, a high level architecture analysis and design is produced. Secondly, different deployment models are investigated, identifying critical factors influencing the cost of the deployment, such as cost of data transfer, runtime operations or maintenance. These alternatives are evaluated through applying them on the concrete use case of a broadcast monitoring system, concluding in quantitative results regarding their efficiency. Utilizing state of the art technologies, such as public cloud computing resources, and taking into account the crucial elements of geographical distribution are also part of the contribution this thesis provides. Lastly, future research directions are proposed