PePeR: A distributed range addressing space for peer-to-peer systems

Abstract

This paper describes a Peer-to-Peer Range (PePeR) addressing space to process the "select" relational algebra operator. PePeR consists of several novel design decisions to support both exact-match and range selection predicates. First, it constructs Z ranges per node in order to efficiently route predicates in a decentralized manner. Second, it employs interleaved range declustering to minimize mean time to data loss in the presence of node removal(s). Third, it uses innovative techniques to adjust its addressing space in the presence of node insertion. The insertion of nodes is done in a distributed manner and we present a technique that approximates a uniform distribution of records across the nodes. In addition, we present performance numbers from PePeR and compare it with a distributed hash table (DHT). The obtained results show the following. If the workload of a relation is dominated by range predicates then PePeR is a superior alternative. On the other hand, if the workload of a relation is primarily exact-match retrievals (selection predicates using the equality comparison operator), then a DHT provides better performance. Both PePeR and DHT may co-exist in a peer-to-peer system simultaneously, enabling a database specialist to employ DHT with one relation and PePeR with another. © Springer-Verlag Berlin Heidelberg 2004.