Prototypes of replicated database management system have been designed and implemented to improve the reliability and availability. Although these systems are quite robust to failure, a recovery mechanism is still missing to bring a failed site back into the system. This thesis studies two different distributed recovery strategies and presents a hybrid distributed recovery algorithm that, combines the two strategies. These recovery algorithms using different strategies are also integrated into PostgreSQL-RR system. This thesis also compares the cost in term of time associated with recoveries using different strategies, which gives a good foundation to the heuristic used in the hybrid distributed recovery algorithm to automatically select an optimal strategy to reduce the recovery time.

This item was digitized as part of a project to share McGill's intellectual legacy with the public. If you are the copyright holder or a relative of the copyright holder who is deceased, you may request withdrawal by emailing escholarship.library@mcgill.ca. The full policy for eScholarship may be accessed here: https://www.mcgill.ca/libraries/research-services/escholarship/policy