This thesis describes the design, implementation and evaluation of a new grid manager for the NorduGrid middleware. NorduGrid is among of the most successful grid computing projects with more than 50 clusters and 5000 CPUs participating. The current NorduGrid grid manager is rigid and difficult to extend with new functionality.

This motivates the design of a new plugin-based event-driven architecture for grid managers. The architecture is implemented in the Python programming language and used as the basis for a new prototype drop-in replacement for the current NorduGrid grid manager. The new grid manager is developed as a drop-in replacement to ease the transition toward supporting novel grid features.

To evaluate the extensibility of the new grid manager, several advanced use cases for the grid manager are examined and approaches for achieving them in the new grid manager are presented. The evaluation shows that the new grid manager is extensible, efficient and customisable. Approaches to improve the fault-tolerance of the new grid manager are presented to improve the resilience to system failures.