A recovery module is one of most important parts in a database management system. When a database system crashes, some data that have not flushed to disk are lost. The module recovers such data loss when a system crashes. Nowadays as a database system is more complex and has more abilities, various transaction services are required. It also has the abilities of transaction services to make a database consistent and durable. This thesis presents the design and implementation of a recovery module for a hypercube database machine. Requirements for the recovery module were set up by analyzing properties of a hypercube database machine called COREDB. Based on the requirements specification was designed based on the requirements by using the FSM (Finite State Machine) formalism. The WAL(Write Ahead Log) method was employed in implementation of the module as a recovery method. To reduce restart time, we maintained a log volume for each database volume. The fuzzy checkpoint method was adapted for efficiency in the normal operation : partial undo ability was implemented to various transaction services. Testing of the recovery module was done successfully in two phases : verification and validation. Execution time of the database system with the recovery module was measured and compared with one without such module.
