Distributed Pattern Matching for P2P Systems

Abstract

Flexibility and efficiency are the prime requirements for any P2P search mechanism. Existing P2P systems do not seem to provide satisfactory solution for achieving these two conflicting goals. Unstructured search protocols (as adopted in Gnutella and FastTrack), provide search flexibility but exhibit poor performance characteristics. Structured search techniques (mostly distributed hash table (DHT)-based), on the other hand, can efficiently route queries to target peers but support exact-match queries only. In this paper we present a novel P2P system, called distributed pattern matching system (DPMS), for enabling flexible and efficient search. Distributed pattern matching can be used to solve problems like wildcard searching (for file-sharing P2P systems), partial service description matching (for service discovery systems) etc. DPMS uses a hierarchy of indexing peers for disseminating advertised patterns. Patterns are aggregated and replicated at each level along the hierarchy. Replication improves availability and resilience to peer failure, and aggregation reduces storage overhead. An advertised pattern can be discovered using any subset of its 1-bits; this allows inexact matching and queries in conjunctive normal form. Search complexity (i.e., the number of peers to be probed) in DPMS is O (log N + zetalog N/log N), where N is the total number of peers and zeta is proportional to the number of matches, required in a search result. The impact of churn problem is less severe in DPMS than DHT-based systems. Moreover, DPMS provides guarantee on search completeness for moderately stable networks. We demonstrate the effectiveness of DPMS using mathematical analysis and simulation results