《ATLaS A Native Extension of SQL》.pdf

ATLaS: A Native Extension of SQL for Data Mining Haixun Wang Carlo Zaniolo IBM T. J. Watson Research Center Computer Science Dept, UCLA Hawthorne, NY 10532 Los Angeles, CA 90095 haixun@ zaniolo@ Abstract fined in SQL itself—rather than in the external procedural A lack of power and extensibility in their query languages has languages required for defining new functions in current O- seriously limited the generality of DBMSs and hampered their R systems. This closure property is the source of great power ability to support data mining applications. Thus, there is a and flexibility, and it is also conducive to a stream-oriented pressing need for more general mechanisms for extending DBMSs computation model. The ability of expressing data mining to support efficiently database-centric data mining appliacations. functions in SQL involves a modest performance overhead To satisfy this need, we propose a new extensibility mechanism for over the same applications written in C/C++ (and an im- SQL-compliant DBMSs, and demonstrate its power in supporting provement over Java or PL/SQL). We prove this by an AT- decision support applications. The key extension is the ability of LaS implementation of the Apriori algorithm, i.e., by apply- defining new table functions and aggregate functions in SQL— ing the litmus test proposed in [1] for database-centric data rather than in external procedural languages as Object-Relational mining. (O-R) DBMSs currently do. This simple extension turns SQL into Besides being particulary eff