Mapping the Gnutella Network Macroscopic Properties of Large Scale P2P Systems Ramaswamy N.Vadivelu Scalab, ASU 03/19/02 Scalab Seminar Series 1 Introduction •Martin Ripeanu, Ian Foster •University of Chicago Gnutella: A Quantitative Evaluation Aims to •Analyze the overlay network topology. •Evaluate the generated network traffic. A P2P system builds a virtual network at the application level. 03/19/02 Scalab Seminar Series 2 •Topology of this network impacts the application. •Affects: communication,performance,reliability,scalability & anonymity •Defining Emergent properties. •Gnutella as a Power Law Network. •Mapping the virtual network to the physical network. •Efficient usage of underlying resources. 03/19/02 Scalab Seminar Series 3 Some issues to be kept in mind Gnutella Protocol: •Servents, •Broadcasting, •Ping,Pong,Query and File Transfer •Summary 03/19/02 Scalab Seminar Series 4 Collecting Data • Gnutella Crawler. •Initiation. •List. •Contact. •Crawling Strategy. •Trade off between discovery time and invasiveness. 03/19/02 Scalab Seminar Series 5 Analysis Network Characteristics: •A Power Law Network •N = L –k, K ; Network constant •Vulnerable to well planned attacks. 03/19/02 Scalab Seminar Series 6 •Small number of devoted users. Traffic Estimation: •Eavesdropping using the Crawler. •Analyzing the distribution of node-node shortest paths. 03/19/02 Scalab Seminar Series 7 •Most links support similar traffic. •Total traffic is proportional to no.of connections. •170,000 connections for a 50,000 node Network •6Kbps per connection. •330 TB/month or 1.7% of total traffic in US Internet backbone(as on 12/00). 03/19/02 Scalab Seminar Series 8 •Focus should be on employing available networking resources efficiently. •Store and forward -> Stress on the infrastructure •Topology mismatch: Experiment •Internet as a collection of AS’s. •AS’s are collections of LAN’s. 03/19/02 Scalab Seminar Series 9 •Crossing AS borders is more expensive than local traffic. •2-5% of Gnu nodes are within the same AS. •>40% of nodes are located within the top 10 AS’s. •-> Most Gnu traffic crosses AS borders. 03/19/02 Scalab Seminar Series 10 Summary •Gnutella Network has features of a Power Law Networks. •Growth of Gnutella dependent on network usage •Two areas of further approach •More efficient P2P designs. •Smarter routing/communication mechanisms. 03/19/02 Scalab Seminar Series 11