David Manura

2002-01-21

CSE-498 (Adv. Networks)

Comparing Hybrid Peer-to-Peer Systems—A Review

The paper “Comparing Hybrid Peer-to-Peer Systems”¹ provides a solid foundation for further research into peer-to-peer file sharing networks and hence is recommended for publication.

The authors develop verifiable models for user behavior and resource consumption, a variety of hybrid peer-to-peer file sharing architectures are presented based on combinations of data locality and of refresh policies for library metadata, and equations are developed to compare the maximum users per server (not the query response time) of each architecture as a function of network usage and resource availability. Though no architecture is completely discounted, batch update login strategies are found to be more scalable than incremental ones, and the chained architecture is judged more appropriate than the unchained architecture used by Napster. For demonstration purposes the models are fitted to data on OpenNap, Nevertheless, the use behavior models developed is inherently general, and the authors explore the effect of positive, negative, and absent correlation between the query probability distribution (indicating which queries users like to submit) and the query selection distribution (indicating which files users like to store).

The research has importance. The obvious consequence is on the design of file sharing architectures for peer-to-peer music sharing applications, such could describe Napster, Gnutella, and Freenet. The query model and the proposed architectures can be applied to P2P file sharing in general, and the research is related, and possible applicable in part, to more diverse P2P or hybrid P2P applications such as cooperative web caching.

The author acknowledges that query response time is neglected in the analysis, but this is a major issue. The chained architecture could be extended in ways that improve query response time. Given the corpa, if a low number of results are found locally, then a relatively low number of remote results can be expected, in which case sending parallel, rather than serial, queries to remote servers could be done to advantage.

Technically, the paper is clear, thorough, and rich in content. However, for space restrictions, the paper offloads some very relevant information (such as how experiments are performed) to a larger technical paper.

The paper is relevant and well written. It is expected that other others may be able to build upon the general models presented.

[1] Yang, B. and Garcia-Molia, H. “Comparing Hybrid Peer-to-Peer Systems.” In Proceedings of the 27^th VLDB Conference, 2001. http://citeseer.nj.nec.com/yang01comparing.html