Generalized bitmap indexes for multi-way equijoin query processing

Historically, full selectivity reduction of join queries has been considered a desirable goal in distributed database management systems. Query selectivity determines which subset of tuples participates in the result. Full reduction refers to precomputing the query using metadata such as indexes before materializing the result tuples. This paper presents a full generalization of bitmapped indexes, showing that: 1) When represented as bit matrixes across extant domains, selectivity indexes and join indexes are simply the transpose of each other. 2) Full selectivity in a multi-way equijoin query can be propagated using standard matrix operations on the indexes. 3) The matrix results can be translated into a non-matrix representation which is more economical of storage space, and for which suitable algorithms, analogous to the matrix operations, achieve the same result of full query reduction based on selectivity. Further work remains in deter-mining the efficiency of the representation and algorithms, integrating non-join selectivities into the reduction, and estimating the maintenance cost for the data structures.