Federated database systems for managing distributed, heterogeneous, and autonomous databases
ACM Computing Surveys (CSUR) - Special issue on heterogeneous databases
| Hi-index | 0.00 |
A heterogeneous distributed database is a system of interconnected DBMSs that use different strategies for data and transaction management. Though issues such as universal query languages, and global view and schema integration have been investigated, transaction management issues introduced by the integration of separate database management systems into one global database have not been widely addressed in the literature.Probably most disturbing to me is the general misunderstanding in the database community as to what a heterogeneous database really is. Many people seem to refer to distributed databases as heterogeneous databases.Of interest to me are questions relating to transaction support in the heterogeneous database environment. Two basic approaches are possible in order to integrate differing DBMS's. The first approach integrates transaction management policies into one global transaction manager that handles subtransactions accessing the heterogeneous database. The second is based on hierarchical composition of transaction management policies. In the latter approach, software is added on top of existing systems whenever needed (Gligor and Luckenbaugh, Interconnecting Heterogeneous Database Management Systems, IEEE Computer January 1984).The heterogeneous database research group at Penn State consists of three students along with myself. We are looking at concurrency control (Y. Leu), recovery (D. Mannai), and deadlock issues (I. Mahgoub). In this panel I would like to discuss the following set of related questions:Which of the problems are due to DBMS integration and not due to heterogeneity?How is the consistency of heterogeneous databases defined?How does serializability apply to heterogeneous databases?How strict must the definition of database autonomy be? How does it complicate transaction management issues?How important are atomic updates in this environment? How often are they expected?Are we likely to have generalized solutions to the problem of concurrency control and recovery?In the discussion we would like to consider all possible systems depicted in the figure below. Especially in the shaded areas.