Posse: A Framework for optimizing incremental view maintenance atdata warehouses

Incremental maintenance of materialized views consists of installing changes tothe view as a result of an update to one of the base relations. This generallyrequires information from some or all of the other base relations comprisingthe view, which may be obtained by probing (i.e. querying) the baserelations. In a distributed context such as a data warehouse, there may be agreat deal of concurrency among updates as well as probes and replies, leadingto consistency problems, for which a number of solutions have been offered.These solutions, however, have been expressed as algorithms of limitedpractical use because they are variously limited to restricted view queries,relational semantics, and limited concurrency. We propose the POSSE frameworkin which many of these limitations are removed, and in which the techniques ofdistributed query optimization can be brought to bear on the problem. To thisend, we show how for a particular method of consistent probing it is possibleto have the power of SQL view queries with multiset semantics, and at the sametime have available a spectrum of concurrency from none at all as in somepreviously proposed solutions to the maximum concurrency obtained by issuingall probes in parallel. We then show how optimization of the probing processcan be used to select various degrees of concurrency for the desired tradeoffsof concurrency against processing cost and message size.