Database management systems for statistical and scientific applications: Are commercially available DBMS good enough?

Abstract

Commercial DBMS are in general not good enough for statistical and scientific applications, but neither are the special-purpose systems that have currently been developed.

Commercially available so-called geographical information systems (GIS) enjoy a commercial attention that other kinds of SSDBMS have never experienced and provide an interesting example. These systems are frequently built on top of a CDBMS, relational or otherwise, but their user language constructs are invariably as ad hoc with respect to the geometric concepts which underlie most operations on geographical data as were the first commercial database management systems with respect to set operations.

Present systems for statistical and scientific database management are thus in their conceptual infancy and much research has to be carried out to improve the situation. Unfortunately, few scientists seem to think that this is a worthwhile research area. Instead, what development there is, seems to be driven mostly by market forces. These forces will rarely favour investments in research and development for science's own benefit.

Storage structures and associated access algorithms, specially developed for SSDBM systems, can provide significant performance improvement for problems occurring frequently in data analysis applications, but only the user can properly assess the value of such improvements in relation to other advantages that the use of a CDBMS may entail.