Krishna P. Gummadi, Stefan Saroiu and Steven D. Gribble

1
Krishna P. Gummadi, Stefan Saroiu and Steven D.
Gribble U. of Washington
2. Measurement Methodology
1. Motivation

• Lots of research and industrial excitement
• Chord MIT, Tapestry,CAN UCB, Jxta SUN,
  Herald, Past MSR, Publius ATT
• A distributed infrastructure largely comprised of
  voluntary, dynamic ad-hoc membership by peers.
• Peers have symmetric roles (serving, downloading
  and routing) throughout system.
• No knowledge regarding the fundamental
  characteristics of peers participating in the
  network
• This knowledge can help in evaluating the
  effectiveness of different schemes.

• Our measurements proceeded in three stages
• Periodically crawl Napster and Gnutella
• discover peers, IPs. overlay topology, and
  whatever metadata about peers
• Feed output from crawl into custom measurement
  tools
• measure bottleneck bandwidth to/from peers using
  SProbe.
• measure IP latency to/from peers
• track content and degree of sharing, where
  possible
• Sub-sample population to measure lifetime
• Track availability of peers at application and IP
  level

3. Results
How many peers have server-like behavior ?
High upstream bandwidth ?
Low latency ?
High availability ?

• Majority of the peers (gt50) connect through
  Cable or DSL modems.
• On average, peers have low upstream bandwidths
  compared to downstream bandwidths, a feature more
  representative of a client than a server.

• Large variation in IP level latencies. For a
  fraction of peers (20) transmission delay ltlt
  latency, implying congestion.
• Session durations strikingly similar in both
  systems. Median session 60 mins. Hence, content
  on a peer unlikely to be available without
  replication.

What is the extent of free-riding ?
How many peers lie about their bandwidth ?

• A large fraction of peers (25) choose not to
  report their bandwidth, they are either unaware
  of it of have no incentive to report it.
• Peers have an incentive to report lower
  bandwidths, a significant fraction do so.
• Lack of knowledge is universal.

• Modem (lt64 Kbps) users share less files and do
  more downloads compared to broad band users.
• Sharing less files Top 7 of nodes share more
  than bottom 75 in Gnutella. 40-60 of peers in
  Napster share only 10-15 of files.

4. Conclusions
5. Future Work

• Apply the results of these measurements to
  evaluate several proposed distributed index
  systems.
• Analyze content life time patterns, and
  geographical distribution of the content and
  peers

• Peers characteristics are very heterogeneous. A
  system should delegate responsibilities to its
  peers based on their characteristics.
• The system should measure the characteristics of
  a peer rather than rely on self-reports from the
  peers themselves.

6. More Information