Vertically-partitioned parallel signature file method

Recently, parallel signature file methods have been proposed for better ret rieval performance in signature files. In this paper, we propose a vertical ly-partitioned parallel signature file (VPSF) method which can partition a signature file vertically. Our VPSF method uses an extendable hashing techn ique for dynamic environment and uses a frame-sliced signature file techniq ue for efficient retrieval. Our VPSF method also can eliminate the data ske w and the execution skew by allocating each frame to a processing node. To prove the efficiency of our VPSF method, we compare its performance with th ose of the conventional parallel signature file methods, i.e., HPSF and Ham ming filter, in terms of retrieval time, storage overhead, and insertion ti me. The experimental result shows that our VPSF achieves about 40% better r etrieval performance than the Hamming filter. In addition, we evaluate the performance of our VPSF methods on several normal distributions with half a nd double standard deviations of the real data. From the performance evalua tion on record sets with half standard deviation, we show that our VPSF gai ns about 20-50% improvement in retrieval time, compared with the Hamming fi lter and the HPSF. Finally, we show that our VPSF generally outperforms the conventional parallel signature files on retrieval performance when the re cords of a database are uniform in size. (C) 2000 Elsevier Science B.V. All rights reserved.