On the Importance of Tuning in Incremental View Maintenance: AnExperience Case Study

We originally set out to explore the performance of a particular algorithm forincremental maintenance of a materialized view in a data warehouse. We choseto use a substantial database schema and population, derived from the TPC-Dfamily of benchmarks, with a population size of around one gigabyte. In theprocess, we discovered that at this size, unexpected issues with the query andupdate processing had to be understood and dealt with before the results wouldbe meaningful. We discover that incremental maintenance is feasible over awide range of update sizes (granularities), and that in all cases acursor-based version of the algorithm performs the best. This is a report ofboth the results of the original inquiry and of some of the eventsencountered.