As views in a data warehouse become more complex, the view maintenance
process can become very complicated and potentially very inefficient.
Storing auxiliary views in the warehouse can reduce the complexity and
improve the efficiency of view maintenance, and the same auxiliary views can
help in efficiently answering lineage tracing queries over the warehouse
views.  In this paper, we study the problem of selecting auxiliary views to
materialize in order to minimize the total view maintenance and lineage
tracing cost. We consider relational views with arbitrary use of aggregation
operators, and we define an initial search space for our optimization problem
based on a normal form for such view definitions.  We present several
auxiliary view selection algorithms, and to study their performance  
we conduct experiments using the TPC-D benchmark in addition to synthetic 
view definitions and statistics. The results of our experiments show: 
(1) the exhaustive algorithm that selects the optimal set of auxiliary 
views is far too expensive in many cases;  
(2) two heuristic algorithms that we present select good (often optimal) 
sets of auxiliary views in a much shorter time;  
(3) even auxiliary views selected by a very simple algorithm can 
significantly reduce the overall view maintenance and lineage tracing cost.