Compressed positionally encoded record filters in distributed query processing.
Date of Award
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Different from a centralized database system, distributed query processing involves data transmission among distributed sites, which makes reducing transmission cost a major goal for distributed query optimization. A Positionally Encoded Record Filter (PERF) has attracted research attention as a cost-effective operator to reduce transmission cost. A PERF is a bit array generated by relation tuple scan order instead of hashing, so that it inherits the same compact size benefit as a Bloom filter while suffering no loss of join information caused by hash collisions. Our proposed algorithm PERF_C (Compressed PERF) further reduces the transmission cost in algorithm PERF by compressing both the join attributes and the corresponding PERF filters using arithmetic coding. We prove by time complexity analysis that compression is more efficient than sorting, which was proposed by earlier research to remove duplicates in algorithm PERF. Through the experiments on our synthetic testbed with 36 types of distributed queries, algorithm PERF_C effectively reduces the transmission cost with a cost reduction ratio of 62%--77% over IFS. And PERF_C outperforms PERF with a gain of 16%--36% in cost reduction ratio. A new metric to measure the compression speed in bits per second, "compression bps", is defined as a guideline to decide when compression is beneficial. When compression overhead is considered, compression is beneficial only if compression bps is faster than data transfer speed. Tested on both randomly generated and specially designed distributed queries, number of join attributes, size of join attributes and relations, level of duplications are identified to be critical database factors affecting compression. Tested under three typical real computing platforms, compression bps is measured over a wide range of data size and falls in the range from 4M b/s to 9M b/s. Compared to the present relatively slow data transfer rate over Internet, compression is found to be an effective means of reducing transmission cost in distributed query processing. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis2004 .Z565. Source: Masters Abstracts International, Volume: 43-01, page: 0249. Adviser: J. Morrissey. Thesis (M.Sc.)--University of Windsor (Canada), 2004.
Zhou, Ying (Joy)., "Compressed positionally encoded record filters in distributed query processing." (2004). Electronic Theses and Dissertations. 1506.