On storage trends and query evaluation

There have been significant improvements in storage technologies over the past few decades. While storage costs (both disk and main memory prices) are decreasing in accordance to Moore's Law (this means a factor of 100 every decade), disk I/O performance is hardly keeping up. In this thesis, we study how query processing can evolve to meet this challenge; in particular how resources like disk space and main memory can be leveraged in order to conserve I/Os. The first part of the thesis proposes a new mirroring strategy termed as fractured mirrors which uses two different storage models to store the mirrored copies. This scheme combines the best aspects of both these models along with the added benefit of mirroring to better serve an ad-hoc query workload. The benefits of this approach are illustrated using experimental results from a prototype system. The second part of the thesis examines how increasing main memory sizes are likely to impact query evaluation. Current query optimizers assume all data is disk resident while optimizing queries. As a result of this assumption the optimizer is likely to miss much better plans (especially with large buffer pool configurations). We examine the benefits of a buffer pool aware query optimizer and propose a generic architecture for the same.