Information Technology Reference
In-Depth Information
An Efficient Lookup Service in DHT
Based Communication System
Kai Shuang, Peng Zhang, and Sen Su
State Key Laboratory of Networking & Switching Technology
Beijing University of Posts & Telecommunications (BUPT)
Beijing, China
{shuangk,susen}@bupt.edu.cn, zhangppmmeer@163.com
Abstract.
Telecom systems utilize the Distributed Hash Tables (DHTs) ap-
proach to build the network infrastructure for advantages of even distribution of
workload, high scalability and cost-effectiveness. Although DHT is undoubted-
ly applicative in such architectures, some practical distinctions still should be
considered to meet the performance requirements of telecom infrastructures.
This paper focuses in two features of the distributed telecom system, so-called
the real-time response and geographic partition, proposes a hierarchical DHT
lookup service named Comb. Comb's overlay is organized as a two-layered ar-
chitecture, workload is distributed evenly among nodes and most queries can be
routed in no more than two hops. Comb performs effectively with low band-
width consumption and satisfactory fault tolerance even in a continuously
changing environment. Both theoretical analysis and experimental result dem-
onstrate that the two-layered architecture of Comb is feasible and efficient.
Comb improves the performances on routing delay and lookup failure rates with
high scalability and availability.
Keywords:
Distributed Communication System, Peer-to-Peer, Comb, DHT,
Two-Hop.
1
Introduction
Distributed Hash Table (DHT) are widely applied in building large-scale self-
organizing overlay networks [1], such as file-sharing, search engines and content
distribution. Currently, both distributed IP Multimedia Subsystem and P2PSIP are
proposed to adopt the structured P2P overlay in communication systems. A funda-
mental problem for a DHT lookup service is resources locating and routing. Overlay
topology, routing path latency and maintenance cost are three elements that impact
the efficiency of a DHT algorithm [1]. Designs of DHT algorithms vary largely, so far
as we know, most proposed DHT algorithms vary routing tables' size from
ON
(log
)
ON
to
(1
O
[2]. Large routing
tables are expensive in maintenance and hard to scale to large systems while long
routing hops lead a long time routing delay. Trade-off should be made between
routing table's size and routing hops in selection of a DHT algorithm.
ON
, with routing hops ranged from
()
(log
)
to
