Novel algorithms for computing medians and other quantiles of disk-resident data

In data warehousing applications, numerous OLAP queries involve the processing of holistic operations such as computing the "top N", median, etc. Efficient implementations of these operations are hard to come by. Several algorithms have been proposed in the literature that estimate various quantiles of disk-resident data. Two such recent algorithms are based on sampling. The authors present two novel and efficient quantiling algorithms, Deterministic Bucketing (DB) and Randomized Bucketing (RB). We have analyzed the performance of DB and RE and extended the analysis of the sampling done in prior algorithms. We have conducted extensive experiments to compare all these four algorithms. Our experimental data indicate that our new algorithms outperform prior algorithms not only in the overall run time but also in accuracy. The new algorithms can be used either as one-pass algorithms to accurately estimate quantiles or as algorithms for computing the quantiles exactly.