A search filter is used to filter out the items that do not exist in the se
arch space. For a search filter, the false drop probability and the average
testing time are two important factors estimated. Bloom filter and Random
filter are two kinds of well-known search filters proposed by Bloom and by
Wang et al., respectively. Chang and Leu had proved that Random filter does
not guarantee to be superior to Bloom filter. In other words, both Random
and Bloom filters have their own fittest performance conditions. This paper
proposes a new search filter mechanism, Partition filter, which improves t
he above two methods with respect to the two criteria, the false drop proba
bility and the average testing time, by taking the advantages of Bloom and
Random filters. (C) 1999 Elsevier Science Inc. All rights reserved.