Incentives: Difference between revisions

From
Jump to navigation Jump to search
m (Multi-Level-Incentives moved to Incentives in the Maze-Network)
(No difference)

Revision as of 10:02, 30 July 2007

Problem

File-sharing networks like KaZaA and Gnutella have popularized the peer-to-peer (P2P) resource sharing model. Based on the altruistic behaviour of its users, principles of P2P systems sometimes conflict user's rational individuality. The individual's utility function is also influenced by negative factors like e.g. upload costs, limited upload bandwidth, enviousness or egoism. Hence some users use P2P networks without sharing own resources (free-riding).

Incentive systems

Point incentive system

Simple point incentive mechanisms as in e.g. the Maze-network don't prevent users from cheating. Two kinds of user collusion are observed.

  • Pair-wise collusion
Two users mutually exchange large amounts of data to increase their points. This kind of collusion profits from the rule that uploading earns more points that downloading consumes.
  • Spam account collusion
A user registers a number of spam-accounts to download from his main-account. Thus he transfers initial points given to new accounts to his main-account.


Tit-for-Tat

Solving satellite cluster problem requires that peers utilize personalized rankings. A simple approach is the private history based Tit-for-Tat. Peers rank their neighbors by normalized download traffic during a certain time period. Thus, future up- and download traffic is influenced by private experience, giving higher priority to "generous" peers.

Simulation Results

Used in large scaled networks like Maze, Tit-for-Tat suffers from its small coverage. Even with long private history only a tiny percentage of peers is known to a user. Hence the majority are "blind" uploads, which bring a lot of opportunity to free-riders. This encourage peers to use "spam"-accounts than establishing long term trust relations.


Links