Materialised view selection using randomised algorithms

A data warehouse stores historical data for the purpose of answering decision making queries. Such queries are usually exploratory and complex in nature and have a high response time when processed against a continuously growing data warehouse. This response time can be reduced by materialising the views in a data warehouse. All views cannot be materialised due to space constraints. Also, optimal view selection is an NP-complete problem. This paper proposes a randomised view selection two phase optimisation algorithm (VS2POA) that selects the top-T views from a multi-dimensional lattice. VS2POA selects views in two phases wherein, in the first phase, iterative improvement is used to select the best local optimised top-T views. These become the initial set of top-T views for the next phase, which is based on simulated annealing. VS2POA, in comparison to the well known greedy algorithm HRUA, selects comparatively better quality views for higher dimensional datasets.