Enriching volunteer clouds with self-* capabilities

Provisioning, using and maintaing computational resources as services is a hard challenge. On the one hand there is an increasing demand of such services due to the increasing role of software in our society, while on the other hand the amount and variety computational resources is growing due to the pervasiveness of computational devices in our lives. The complexity of such problem can only be mastered by resorting to suitable technologies based on well-studied paradigms. Three prominent examples and ICT trends of the last decade are (i) cloud computing, which promotes the idea of computational resources as services; (ii) autonomic computing, which aims at minimizing the amount of human intervention and automatizing many aspects of a system’s life-cycle; and (iii) volunteer computing, which promotes the idea of achieving complex tasks by fostering the collaboration among peers. This thesis proposes an approach based on the combination of the above mentioned paradigms (i)–(iii) for the design and evaluation of volunteer cloud platforms providing a service for executing simple tasks. The major problem under consideration is the selection of the mechanisms used by cloud participants to collaborate for providing such service. The main contributions of the thesis are: (1) an architecture and a model for volunteer cloud platforms; (2) a discrete event simulator for such model; (3) the extension of a statistical analysis tool to ease the analysis; (4) novel self-* strategies for collaboration among volunteers, mainly inspired by multi-agent systems and AI techniques, evaluated with the simulator using the Google Backend workload.