Foundations for faster external sorting

Abstract

The availability of large main memories and the new technologies for disk drives have modified the models for external sorting and have renewed interest in their study. We investigate the replacement-selection paradigm for external sorting. We focus on the performance of the merge phase when given random files and specially when given nearly sorted files since such files are common in practice. In particular, we demonstrate that, during the merge phase, the floating-buffers technique not only reduces the sorting time by fully overlapping I/O and saving seeks, but also it profits significantly from existing order in the input. We also propose a new algorithm for computing a feasible reading sequence for floating buffers that improves upon previous algorithms.

This work was partially carried out under grants from the Natural Sciences and Engineering Research Council of Canada, the Information Technology Research Centre of Ontario and CONACYT 1707-A9209 in México.