The Hybrid Technique for Reference Materialization in Object Query Processing

Resolving object references, or reference materialization, is a fundamental operation in object query evaluation. Existing reference materialization techniques fall into two categories: pointer-based and value-based. We identify several drawbacks of existing techniques, and propose a hybrid technique that combines the advantages of each category. This technique relaxes the limitations of valuebased techniques, while preserving much of their performance advantage over pointer-based techniques; it performs well in the cases when no existing algorithm is applicable or efficient. The hybrid technique shows even stronger performance advantages when moving from single-valued to collection-valued attributes. We present algebraic transformations to enable the hybrid technique in the rule-based query optimizer. Initial experimental results using a commercial object-oriented database show that the hybrid approach achieves significant speedup over current algorithms in many cases. The initial motivation for our work was optimization and evaluation of object-oriented query languages, particularly OQL [ODMG]. However, the key features we have concentrated on, references and collectionvalued attributes, are present in object-relational products and the SQL:1999 proposal [EM99].