presentation of article:

1

• presentation of article
• Small-World File-Sharing Communities
• Article Adriana Iamnitchi, Matei Ripeanu, Ian
  Foster
• Presentation Periklis Akritidis
• ICS-FORTH

2
Patterns in file-sharing communities

• Small-world patterns exist in diverse
  file-sharing communities.
• A high-energy physics collaboration
• The Web as seen from the Boeing traces
• The Kazaa peer-to-peer file-sharing system
• Motivation can be exploited for mechanism design.

3
Data Sharing Graph

• A graph in which nodes are users and an edge
  connects two users with similar interests in
  data.
• Similarity criterion the number of shared
  requests within a specified time interval
• Degrees of freedom
• length of time interval
• threshold on the number of common requests

4
Small-World Characteristics of Data Sharing Graph

• Clustering Coefficient (see article for
  definition)
• Large, much larger than that of a random graph
  (Poisson distribution for node degree) with the
  same number of nodes and edges.
• Average Path Length
• Small, like random graph.

5
Methodology

• Compare clustering coefficients and average path
  lengths for various communities with random
  graphs of same size

6
High Energy Physics Collaboration
7
Web
8
Kazaa
9
Possible Bias Large clustering coefficient of
unimodal affiliation networks

• A bipartite network (left) and its unipartite
  projection (right).
• Users A-G access files m-p.
• In the unipartite projection, two users are
  connected if they request the same file.
• The projection is inherently more clustered than
  a random graph.

10
Possible Bias degree distribution

• Non-poisson degree distribution of data-sharing
  graphs may cause small-world characteristics
• Degree distribution was Zipf for Kazaa and Web
• Newman et al. propose a model for random graphs
  with given degree distributions

11
Evaluating Bias

• Compare against the clustering of unimodal
  projections of random affiliation networks of the
  size and degree distributions given by traces.
• Results

12
User-independent trace characteristics

• User-independent characteristics of traces
• event frequency follows Zipf distribution
• time locality
• temporal user activity
• Are the observed patterns an inherent consequence
  of these well-known behaviors?
• They processed the traces preserving the
  documented characteristic but breaking the
  user-to-request association
• The resulting graphs are less small-world
  graphs than their corresponding real ones