Precomputation techniques for interactive data exploration

This thesis explores several complementary subjects in the pursuit of interactive, large-scale data analysis. To be interactive, the database system must precompute large portions of the expected queries; otherwise, the queries are forced to spend too much time accessing large volumes of data. This thesis covers three aspects of precomputation: Iceberg Cube, index selection for slice queries, and asynchronous incremental materialized view maintenance.; We introduce the Iceberg Cube problem, which is to compute an OLAP Cube where every cell must meet a minimum support threshold. The Iceberg Cube vastly reduces the storage requirements of the full Cube, and is an attractive alternative for OLAP view selection techniques. We present an algorithm called Bottom-Up Cube (BUC) that computes Iceberg-Cubes and sparse Cubes significantly faster than previous techniques.; When records that are pruned from the Iceberg Cube must be found (e.g., for incremental maintenance or when the user requires fine-grained details), indexes must be used to locate the records. Choosing the right indexes is challenging because of the enormous number of possible indexes and queries. We formalize the multidimensional index selection problem for slice queries and compare several algorithms that empirically found near-optimal configurations. Once materialized views are added to the DBMS, the views must be incrementally maintained as the database is updated. Incremental view maintenance suffers from serious concurrency problems, especially for Cube views. We describe a technique called point-in-time refresh that reduces the lock contention.; This thesis also describes Dax, a data exploration and visualization system that presents many coordinated views of the data. At the heart of Dax is MQL, a query language extension for describing coordinated views, which is more of a model of database interaction and a technique to include the database system in the Model-View-Controller user interface paradigm than a true query language.