In this thesis, recovery algorithm using the shadow concept is presented, and its performance is evaluated. In the literature, Lorie's shadow page algorithm needs to maintain local copy of page tables while the transactions are processed. Consequently, when the transactions commit, the overhead of merging page tables and the I/O overhead for updating the master record become high. Therefore, the new shadow page recovery algorithm is proposed in this thesis to reduce the time overhead. The new method always maintains shadow pages and current updated versions of page tables, so the method needs more disk space than Lorie's algorithm.
Furthermore, the performances of the two algorithms, Lorie's and the new one, are evaluated and compared for distributed database systems. To do this, a transaction processing model is presented. Results show that when all page tables are maintained in main memory, the algorithm proposed in this thesis performs better than Lorie's algorithm. But when all page tables cannot be maintained in main memory, the average response time of two algorithm mainly is dependent on by transaction interarrival time. When the interarrival time of transaction is short, i.e, the number of transaction processed currently is large, the new algorithm performs better than Lorie's algorithm. However, when the interarrival time of transaction is long, Lorie's algorithm performs better than the new algorithm. If I/O processing time can be reduced, the overhead of the new algorithm could be less than the overhead of Lorie's algorithm.
