Optimization of Boolean expressions for main memory database systems

With the ubiquity of main memory databases which are increasingly replacing the old disk-oriented databases, relations are being stored in denormalized form in order to increase the query throughput, thus, the dominance of join operators in terms of costs is being replaced by the costs of evaluating selection predicates. Boolean expressions containing selection predicates connected both conjunctively and disjunctively have been thus far solved by rather simple heuristics which leaves a large optimization potential unharvested. To exacerbate the matter, such heuristics rely on the independent predicate selectivity assumption which typically does not hold, and the constant predicate costs assumption which in terms of main memory database systems does not hold either. In this thesis we tackle the problem of optimizing Boolean expressions by not relying on the independence assumption nor the constant predicate costs assumption. We present optimization algorithms for queries containing both conjunctively and disjunctively connected predicates together with a cost model which precisely captures CPU architectural characteristics such as branch misprediction. Our optimization algorithms achieve the optimum in terms of plan quality, thus, they harvest the entire optimization potential inherent in Boolean expressions.