Design And Implementation Of Materialized View Of Da Meng Database Management System

Materialized view is an excellent query optimization method used in analytic processing environment. Costly and frequently used queries can be stored as materialized views to answer the future queries. To fulfill the requirements of analytic application context, such as OLAP and data mining, materialized view must be supported by DM. To achieve the support of materialized view, the maintenance of materialized view and how to use materialized to improve query performance, these two key technology problems must be solved properly. Moreover, how to provide an implementation framework according to DM is also should be taken in mind.Under analytic application environment, the modifications happened on base tables are minor compared to the mass of records stored in the whole database. Maintain materialized views incrementally can be much cheaper than recomposing views from scratch over the updated sources. Eight incremental derivation rules of SPJ query and two addition algebra operators are proposed to maintain materialized views of DM incrementally. Correctness of these rules and operators are proved by relational algebra.To ultimate query with materialized views, all possible equivalent rewrites based on materialized views are generated, and the cheapest one was selected as the target query. The whole rewrite procedure is separated into four different sub procedures: the multiply of relations, the condition of selection operation, attributes used in group operation and the output in the selection list. An equivalent rewriting rule is applied in each sub procedure, and the correctness of these rules is proved.The framework of DM database management system is modified to invite in materialized view mechanization. Materialized view is implemented as a new database object, and several DDL statements are created to maintain the Meta data of materialized views in data dictionary. The materialized views are indexed by a multi-pass tree called QRI (Query Rewrite Index).All the possible query rewrites using materialized views are generated by traveling QRI. The cost of each rewrite is calculated by cost estimator. The query plans of data updating statements are modified. When a relation relied by materialized views is to be modified, the modified data is firstly stored in assistant tables, then materialized views and relation is maintained by the original relation and assistant tables.