The case for partial indexes

Current data managers support secondary and/or primary indexes on columns of relations. In this paper we suggest the advantages that result from indexes which contain only some of the possible values in a column of a relation.