Data warehouse maintenance is the task of updating a materialised view to reflect changes to autonomous, heterogeneous and distributed sources. Selection of a maintenance policy has been shown to depend on source and view properties, and on the user specified criteria (such as staleness, response time etc.), which are mapped on to evaluation criteria. In our previous work, we have analysed source and view characteristics, and user requirements to derive a cost- model. Maintenance policy selection has thus been cast as an optimisation prob- lem.
This paper takes a complementary approach to evaluating maintenance policies, by implementing a test-bed which allows us to vary source characteristics and wrapper location. The test-bed is instrumented to allow costs associated with a policy to be measured. An actual DBMS (InterBase) has been used as a relational source and an XML web server has been used as a non-relational source. The experiments clearly show that maintenance policy performance can be highly sensitive to source capabilities, which can therefore significantly affect policy selection. They have further substantiated some of the conjectures found in the literature. Some of the lessons learnt from this test-bed implementation and eval- uation are reviewed.