This thesis describes the design of U.P., a decentralised job distribution system for mass computing projects. The design is based on the concept of distributed hash tables, which when combined with keyword-based indexing and load balancing provides a robust and scalable peer-to-peer network for mass computing projects to distribute jobs.

An implementation of the system is developed in C++ for the p2psim simulator. Realistic and synthetic tests of the implementation are conducted to explore the intrusiveness, scalability and robustness, with promising results, although a few problems remain.

Hence, we conclude that a decentralised architecture can be a viable alternative to the currently deployed centralised designs with better scalability, high availability and easier and cheaper deployment for new projects. We also believe that the decentralised architecture can be used to provide the foundation of a more general computational grid.