High-Performance Extensible Indexing

Today’s object-relational DBMSs (ORDBMSs) are designed to support novel application domains by providing an extensible architecture, supplemented by domain-specific database extensions supplied by external vendors. An important aspect of ORDBMSs is support for extensible indexing, which allows the core database server to be extended with external access methods (AMs). This paper describes a new approach to extensible indexing implemented in Informix Dynamic Server with Universal Data Option (IDS/UDO). The approach is is based on the generalized search tree, or GiST, which is a template index structure for abstract data types that supports an extensible set of queries. GiST encapsulates core database indexing functionality includingsearch, update, concurrency control and recovery, and thereby relieves the external access method (AM) of the burden of dealing with these issues. The IDS/UDO implementation employs a newly designed GiST API that reduces the number of user defined function calls, which are typically expensive to execute, and at the same time makes GiST a more flexible data structure. Experiments show that GiST-based AM extensibilitycan offer substantially better performance than built-in AMs when indexing userdefined data types.