Performance evaluation of signature based multikey access methods

Abstract

In this thesis, we evaluate the performance of signature based multikey access methods and thus provide guidelines for selecting the most effective one to a given operational environment. We first establish a taxonomy of signature base access methods; One-Path Single-Level(OPSL), Two-Path Single-Level(TPSL), One-Path Two-Level(OPTL), Two-Path Two-Level(TPTL). Then we implement them and perform experiments on various environments. We show form the experimental results that two-path methods(TPSL and TPTL) outperform one-path methods(OPSL and OPTL) on retrieval performance with the expense of a modest 3-15\% storage overheads. But the retrieval performance of single-level methods and two-level methods is subject to the size of data file. When a data file contains 10,000 records, single-level methods show better retrieval performance than two-level methods, while when a data file is composed of 100,000 records, two-level methods outperform single-level methods. Therefore, we recommend TPSL for application systems requiring fast data record retrieval when the data file size is 10,000 records. We also recommend TPTL for most of the interactive systems to achieve fast text retrieval, and OPTL for systems requiring small storage overhead, when the data file size is 100,000 records.