Calibrating the Query Optimizer Cost Model of IRO-DB, an Object-Oriented Federated Database System

Abstract

IRO-DB is an object-oriented federated database system to access multiple data sources from an ODMG compliant C++ interface. The system encompasses several components, including local database adapters to homogenize local data sources, a remote object access component to query and transfer collections of objects from site to site, and a mediator to define integrated views, decompose and optimize queries, and combine results. This paper gives an overview of the IRO-DB architecture and describes in detail the cost evaluator currently under elaboration for the next version of the distributed query optimizer. The cost model is composed of a set of mathematical formulas with coefficients to estimate the cost of the search operators. The coefficients are deduced from a calibrating objectoriented database composed of linked collections of objects. A tuning application is run on each local site to adjust the cost formulas and fix the coefficients. We report on the tuning of O2 and ObjectStore....