PISA: Performance models for Index Structures with and without Aggregated data

Different models to estimate the performance of tree-based index structures exist. Materialized aggregates in the inner nodes of such index structures are used to speed up range queries on aggregates. This is achieved by avoiding traversing the index tree down to the leaves through aggregates precomputed in the inner nodes. None of the existing models deals with these aggregated data. In this paper, we extend the existing models to take account of aggregated data. Our main contribution is a new generic performance model to estimate the Performance of Index Structures with and without Aggregated data (PISA) that covers both aspects. In addition, the PISA model is adaptable to the distribution of the data and to the location of the query boxes (e.g. uniform, normal and skewed distributions). Experiments confirm that, in most cases, the PISA model is more accurate than other models.