Scheduling the allocation of data fragments in a distributed database environment: a machine learning approach

Different database fragmentation and allocation strategies have been proposed to partially replicate data in a partitioned, distributed database (DDB) environment. The replication strategies include database snapshots, materialized views, and quasi-copies. These strategies are `static´ and do not adapt to the changes in the data usage patterns. Furthermore, they often require expensive update synchronizations to maintain data consistency and do not exploit the knowledge embedded in the query history. This paper describes a machine learning based time invariant fragmentation method (MLTIF) that acquires knowledge about the data usage patterns for each node. Based on this knowledge, MLTIF designs time invariant fragments and schedules its allocation and selective update for a specified time period. Simulation is used to compare the effectiveness of the MLTIF approach with that of full replication, materialized views, and nonreplication strategies. Initial results indicate that for most normal operating conditions, the MLTIF approach can be effective