Transcript Sahasrabudhe

Review of
Free Riding on Gnutella
Eytan Adar and Bernardo Huberman
Shreeram Sahasrabudhe
Jargon
Gnutella
- Protocol created by
nullsoft guys (Winamp),
network followed
- Decentralized and Peerto-Peer
- Anonymity maintained
- Limewire, Morpheus
etc. use this protocol
Free Riding
1> Peers that only
download files
without ever
providing files for
others
2> Peers providing
files but not
contributing to the
desirable content on
the network
What it [paper] speaks?
• Presents an extensive experimental analysis to illustrate
and caution us about the effects of free riding in the
Gnutella Network.
• Argues that Free Riding leads to
– System Degradation
– Adds Vulnerability to the system
• First proves that Free Riding does occur
• Gives three hypothesis (confirms by experiments)
1. A significant portion of Gnutella peers are free riders
2. Free riders are distributed evenly across different domains
(and by bandwidth)
3. Peers that provide files for download are not necessarily those
from which files are downloaded
Experiment
• Setup
– Java based Furi client [Fu00] (A Gnutella
Client) was modified to log messages flowing
through.
– Executed for 24hr period
– Logged the ping (are you there?), pong (IP,
port, # of files shared and the total size) and
Query Response Messages (IP, port, ClientID,
other location info)
– Also considers effects NAT based peers some
of which can report the same address.
– 33,335 hosts sharing 3,100,464 files, 87,668
query response messages
Experiment 1
(Free Riding)
• Approximately 69% (inc NAT based peers) of the peers
share no files and 73% share ten or less files.
• Top 1% represent approximately 37 percent of the total
files shared. This quickly escalates to the top 20%
sharing 98% of the files.
• 63% of those sharing files never provided a query
response. Again top 1% provides 47% answers.
Experiment 2
(Free Distribution)
TLD (top-level
domain e.g
(edu, net, org..)
• Domains can function as a proxy for bandwidth
• Filtered the test set to 26,014 peers with IP hostnames
• Peer count linearly related to files shared
Even distribution of free riders
Experiment 3
(“quality” vs. quantity)
• Concentrated queries on particular topics
Number of queries answered is not necessarily proportional
to the number of files offered.
Effect(s)
Network
– Two Factors
Cause:
Ultimately few individuals will contribute anything that is new and high quality.
(The Tragedy of the Digital Commons)
Solutions:
• Uploading files
• Peers participating actively in the protocol
Search Horizon Increases - Hosts can become unreachable
Saturation - Peers have limited number of connections
Vulnerability
• The few providers begin to appear as a centralized server consisting of several
peers.
• Prosecuting Agencies like RIAA can target top-serving peers
• Loss of anonymity
Solution - Free Riding
• FreeNet
– Forces caching of downloaded files
• Users must know -> Identifiers for files
• Can replicate harmful / illegal data and taint hosts
• Napster
– Defaults download directory to be the upload directory
• Spawn – buy & sell; public  private
• Usenet - Reduce the personal storage and BW cost