The HBΞ -tree: a concurrent and recoverable multi-attribute index structure

Abstract

The number of applications that deal with multi-attribute (point or spatial) data is continually increasing. These applications require a Database Management System (DBMS) which offers the same functionality for this kind of data as it offers for traditional data.The DBMS should use effcient and reliable ways to store, index, and access the data. It should also maximize concurrent accessing of the data by as many users as possible at the same time, and be able to recover from application errors or system crashes that result in data inconsistency.Approaches that use multiple single-attribute indexes are quite inefficient. That is why there has been extensive research on explicitly multi-attribute indexing. Most proposed multi-attribute indexes do not offer performance guarantees and well understood methods for concurrency and recovery. But these are the requirements for the inclusion of an index in a general purpose DBMS.We propose a new multi-attribute index. Our approach combines the hB-tree (Lomet & Salzberg), a multi-attribute index with promising performance guarantees, and the Pi-tree (Lomet & Salzberg), an abstract index which offers well understood and efficient concurrency and recovery methods. We call the resulting method the hB\Pi-tree. We describe several versions of the hB\Pi-tree, each using a different node splitting and index term posting algorithm. We also describe a very efficient new node deletion algorithm.We have implemented all the versions of the hB\Pi-tree. Our performance results show that even the version that offers no performance guarantees, actually performs very well, in terms of storage utilization, index size (fan-out), exact-match and range searching, under various data types and distributions. We have also shown that our index is fairly insensitive to increases in dimension. Thus, it is suitable for indexing spatial (non-point) data that is mapped to higher dimensional points. This property and the fact that all our versions of the hB\Pi-tree guarantee very high concurrency, make the hB\Pi-tree a promising candidate for inclusion in a general purpose DBMS.