Gnutella

Peer-to-Peer Architecture Case Study: Gnutella Network
Matei Ripeanu, U Chicago

[Content]

• Crawl and analyze Gnutella network from 10/2000 - 06/2001.


• For save time, use Client/Server crawler. Single client need 50hr for 4000 nodes.


• Power-law is a general distribution in natural, includes HTTP host connection, molecules in a cell, people in a social group, etc. N = K exp(-C)


• Result:


• 
  
  
  
  • Power-law node-shared file number distribution
  
  
  • Average node connectivity: 3.4
  
  
  • Power-law node connectivity distribution in Nov. 2000, but in Mar. 2001, node with <10 connectivity is constant. >10 power-law distribution.
  
  
  • Node-node shortest paths: Avg: 4.24 (Nov. 2000) -> 5.35 (May 2001), Largest: 12, > 95%: < 7 hops
  
  
  
  

[Questions]

• In del.icio.us, people is connected by their bookmark. Believe the bookmark distribution is power-law. How about the shared bookmark distribution? Same questions for Technorati Tag distribution. If two people have multiple shared Bookmarks/Tag are same, we look them as friends. Its a good method to find friends. How about do a Web tool for it? i.e. Google Friends?


• In Jun 2001, newer version of Gnutella make its overhead traffic cut from 64% to 8%, that's great! How and who do it?

[Futher Reading]

• http://gnutella.wego.com/


• Zipf's distribution: includes incomes, word frequencies, Gnutella query


• 
  
  
  
  • L. Adamic, Zipf, Power-laws, and Pareto - a ranking
    tutorial, online tutorial: http://www.hpl.hp.com/shl/
  
  
  
  


• DSS group's Gnutella research


• 
  
  
  
  • DSS group, Gnutella: To the Bandwidth Barrier and
    Beyond, available at: http://dss.clip2.com, Nov. 6, 2000.
  
  
  • DSS group, Bandwidth Barriers to Gnutella Network
    Scalability, available at: http://dss.clip2.com, Sept. 8, 2000.
  
  
  
  

Technorati : Gnutella, P2P, Power-law
Del.icio.us : p2p