Presentation on theme: "On the Economics of P2P Systems Speaker Coby Fernandess."— Presentation transcript:
On the Economics of P2P Systems Speaker Coby Fernandess
Internet Internet became a major force of change on civilization. The Internet itself is the largest distributed computer system. – Client Server Web browsers servers. Relational database servers. As computer networks evolve into more complex system, there is a growing demand for scalable and reliable system architecture.
Network Games Algorithms, games, and the internet. [Papa STOC01] The most significant characteristic of the internet is its social-economic complexity. The Internet is built, operated, and used by entities with diverse economic interests. – Surfers, ISP, Infrastructure provider., Web companies. The mathematical tools and the insights most appropriate for understanding the Internet may come from game theory.
Peer-to-Peer Systems Connectivity between computers shifts from centralized to highly distributed and symmetric systems. Initiated by the phenomenal popularity of p2p file sharing applications. – Napster, Gnutella and BitTorrent. Introduce pure symmetric (server-less) distributed systems. “web-communities" On the internet, where peers voluntary share resources.
Example - Content Distribution Next Popular File-sharing facilities such as, BitTorrent and Emule. Source may become choked with a “flash crowd”.
Collaborative Content Distribution Now, leave me alone!
Objective Incentives for Cooperation Design and implement distributed systems that encourage user cooperation: – incentive compatible p2p systems; inducing cooperation from strategic peers. I.e. prevent or limit “Free Riders”: – Users who attempt to benefit from the community resources without sharing their own resources. Characteristics: – Peers rational autonomous and unreliable. – No central authority. – Highly dynamic membership, (a.k.a. high turnover)
Monetary payments schemes Service consumers pay the service providers. Incentive compatibility: – Distributed Algorithmic Design Mechanism. Impractical since they require an infrastructure for accounting and payments. – Decentralized payment schemes. – Ad hoc networks.
Trust and Reputation System Trust is required for multiple entities to co- operate and share resources. Otherwise, either parties refuse to interact – increased costs. Trust and reputation play an important role in most disciplines, including sociology, economics and computer science. – Trust is the subjective expectation a node has about another node's future behavior, based on their history – Reputation is the perception a node creates through past actions about its intentions and norms.
Trust-Based Schemes Users can always base their actions on their own personal history – E.g., if A previously cooperated with me, I will now cooperate with A Numerous drawbacks to this approach – Repeated interactions with same peer may be infrequent – System boot-up. – Integration of new users.
Reputation-Based Schemes Information about parties’ past behavior can be disseminated through the network – Decisions about future actions no longer based on personal history alone. Newcomers: (Sybil attack) Desirable for network growth, but allows misbehaving users escape the consequences. – Irreplaceable pseudonyms requires central authority. – Penalty to newcomers. – Collusive behavior.
Robust Incentive Techniques for Peer-to- Peer Networks Generalized Prisoner’s Dilemma (GDP) in client/server encounter. The GPD payoff matrix satisfies: – Mutual cooperation has highest total payoff. – Defection (weakly) dominates cooperation.
Reputation System Use subjective reputation to deal with collusion. Max-flow algorithm works well (constant mean time version which doesn’t scale)
Decision Function Decision Function: – Use client ‘s reputation. – Require converge to cooperation – Robust against defection strategies and collision. – Generosity Cooperate with entity X with probability:
"Third of humanity has not yet made a phone call."