Relational Schema Evolution Management In Information Grid

This paper studies the relational schema evolution problem in information grid. Many applications in information grid integrate distributed, autonomous relational databases with key and foreign key constraints. When the database schema evolves autonomously, users can not get correct information without modifying the applications.We present an EVP address space model for the information grid. The application attribute refers to the virtual information source attribute, which in turn refers to the physical information source attribute. When the schema of the physical information source evolves, we just need to adjust the mapping from physical information source attributes to virtual information source attributes, without modifying the applications. Therefore, the relational schema evolution problem is transformed to the mapping adaptation problem.To solve the mapping adaptation problem, we propose a schema decomposition model. This model decomposes a nested relational schema into a set of atomic schemas, each of them constitutes of a primary key or a primary key plus a non-primary attribute. Then we can set up the mapping between attributes suitable for mapping adaptation.We develop a mapping publishing rule and a mapping publishing algorithm based on the rule. And then we analysis various basic relational schema changes and give the corresponding mapping adaptation algorithms. When a physical information source schema changes, the algorithms adjust mappings automatically and affect the fewest number of mappings. We prove the correctness of the mapping publishing algorithm and the mapping adaptation algorithms. We show that the algorithms compare favorably with related work.We formally prove that the above approach solves the relational schema evolution problem in information grid. Also we implement and validate the approach in Vega railway information grid system. With the approach, the time to adjust system is reduced from days to milliseconds. We analyze the performance of mapping adaptation algorithms in the system. The result shows that the mapping adaptation is not the system performance bottleneck. At last, we present a mapping adaptation approach adaptive for publishing schema changes. The approach saves human labor when schema changes are published.