Review - Physical Database Design for Relational Databases.

Surajit Chaudhuri: Review - Physical Database Design for Relational Databases. ACM SIGMOD Digital Review 2: (2000)

Review

There is an increasing awareness that performance tuning of database systems needs to be automated in order to reduce the total cost of ownership. Among other factors, the performance of a database system depends on a judicious choice of its physical design. Therefore, automating the choice of physical design of databases is an important problem. When I started working on this problem in early 1996, this was one of the papers that influenced my thinking. The paper describes DBDSGN, an experimental physical design tool developed at the IBM San Jose Research Laboratory. The focus of the paper was to automate the process of selecting indexes for a database. Two design decisions make DBDSGN an early pioneer and precursor to the index selection tools of today. First, DBDSGN made the selection of indexes based on the workload experienced by the system. Next, the attractiveness of a set of indexes was measured by the estimated cost of the workload for the given index configuration. Each of these decisions is crucial for the success of an automated physical design tool: (a) A workload must be used since it captures the usage of the database and provides a basis for selection. (b) It is also important to use the optimizer-estimated cost as the metric since the optimizer determines usefulness of an index by invoking the cost estimation module during selection of the execution plan. While the paper do not answer many of the critical system issues involved in building an industry-strength index selection tool, the foresight of the designers of DBDSGN in making these key architectural designs make this paper a classic for research in physical database design.

References

[1]: Sheldon J. Finkelstein, Mario Schkolnick, Paolo Tiberio: Physical Database Design for Relational Databases. ACM Trans. Database Syst. 13(1): 91-128(1988)