Daniele Costante Zilio

Trends in automating physical database design

A database administrator (DBA) must make many decisions about what is the best decision for a database and its users - including what is the best physical DB design decision. The DBA must determine what auxiliary data structures to create - such as indexes and materialized views - as well as the optimal way to partition each table in a shared-nothing partitioned system, to maximize workload performance while minimizing any increase in disk space or maintenance costs. Given the number of possible configurations, the possible interactions between each choice, and the associated constraints on the decision, the problem of selecting an optimal design is daunting for even the most skilled DBA. It has long been recognized that DBAs need tools to help them make these complex decisions, but until recently little progress was made. We present our vision for the future of automating physical DB design, and contrast it with previous attempts at solving this problem.

Poster Session Problem definition for effective workload management

The paper introduces the problem of designing dynamic workload management (WM) tools that are aware of the diversity of classes of users and their diverse access patterns. Our approach should be contrasted with the current WM tools and their ability of detecting performance degradations in accordance with the users preference goals. We define the problem and suggest a formal definition that allows the further development of algorithms and architectures that allow the implementation of effective on-line database tuning strategies.