A well-known challenge in data warehousing is the efficient
incremental maintenance of warehouse data in the presence of source
data updates.  In this paper, we identify several critical data
representation and algorithmic choices that must be made when
developing the machinery of an incrementally maintained data
warehouse.  For each decision area, we identify various alternatives
and evaluate them through extensive experiments.  We show that picking
the right alternative can lead to dramatic performance gains, and we
propose guidelines for making the right decisions under different
scenarios.  All of the issues addressed in this paper arose in our
development of WHIPS, a prototype data warehousing system supporting
incremental maintenance.