Flexible Bloom Filters for Searching Textual Objects

Efficient object searching mechanisms are essential in large-scale networks. Many studies have been done on distributed hash tables (DHTs), which are a kind of peer-to-peer system. In DHT networks, we can certainly get the desired objects if they exist. However, multi-word searches generate much communication traffic. Many studies have tried to reduce this traffic by using bloom filters, which are space-efficient probabilistic data structures. In using such filters, all nodes in a DHT must share their false positive rate parameter. However, the best false positive rate differs from one node to another. In this paper, we provide a method of determining the best false positive rate, and we use a new filter called a flexible bloom filter, to which each node can set the approximately best false positive rate. Experiments showed that the flexible bloom filter was able to greatly reduce the traffic.