This VLDB-99 industrial paper describes the Informix commercialization of the Berkeley GIST indexing method. This work offers some significant improvements on the original GIST scheme; the improvements are based on a nicely engineered new extensibility API that greatly reduces the number of user-defined function calls that are needed for a given GIST-based indexing method to get its various jobs (i.e., lookup, insert, etc.) done. The paper even includes a performance comparison between R-trees implemented using "new improved GIST" and Informix's own hard-coded R-tree access method -- interestingly, the GIST-based R-tree version beats its hard-coded competitor.
The significance of this paper, to me, is that before reading it I thought GIST was cute but primarily of academic interest due to its non-optimized nature as originally presented. This paper shows that GIST can indeed be made very practical, and therefore that it's worth doing. It also explains how Informix has actually implemented it, and it includes a brief but real performance comparison. This struck me as a neat industrial paper that also makes a contribution to the state of the art that academics should be very interested in. It's also a very well-written paper, by the way.
Copyright © 2000 by the author(s). Review published with permission.