Lecture Notes for Client-Server Interfaces

17 April 2003 - Peer-to-Peer Systems

Outline

• Client-server systems.

• Peer-to-Peer systems.

  • Characteristics, problems, and issues.

• Peer-to-peer systems.

  • Napster

  • Gnutella

  • Freenet

Client-Server Systems

• Each process acts as a client or a server.

  • Occasionally a process may change, but usually only briefly.

• The client and server roles are distinct.

  • Clients make requests; servers reply to requests.

Client-Server Problems

• Only servers reply.

  • Leads to bottleneck and reliability problems.

• Distributing and replicating the servers is not easy.

  • A problem with distributed systems, not just client-server systems.

• A generally static server architecture.

  • Unlike clients, servers don't normally come and go.

• An impoverished discovery architecture.

  • Usually a fixed list of alternate server addresses.

  • Enriching discovery is a the middle-ware issue.

Peer-to-Peer Systems

• Consider information storage and delivery.

  • Consistent with current practice; other functions are possible.

• Combine the client and server roles into a single peer role.

  • Each peer process acts as a client and a server.

Peer-to-Peer Characteristics

• Any peer has the potential to respond.

  • Eliminates the bottleneck and reliability problems.

• A highly, and easily, distributed system.

  • A problem with distributes systems, not just client-server systems.

• A dynamic and mostly ad-hoc system architecture.

  • Peers may enter and leave arbitrarily.

• Simplified discovery.

  • Put your request in bottle and throw it in the ocean.

Peer-to-Peer Problems

• Uncertainty (probabilistic results).

  • Maybe the answer's out there, maybe not.

  • Retrieval characteristics may change from one request to the next.

• Excessive resource requirements.

  • Storage redundancy, network traffic.

• Protocols more complex than client-server protocols.

  • Protocols have to deal with uncertainty.

• Social problems.

  • The free-rider problem (the Tragedy of the Commons).

  • Antisocial behavior, illegal behavior.

Comparison

• Partially vs. fully decentralized autonomy.

  • Set-up and use.

• Static vs. dynamic architectures.

  • Externally dynamic (processes leave and enter).

  • Internally dynamic (optimizing reconfigurations).

Peer-to-Peer Issues

• Peer discovery.

  • A problem when peers can come and go at will.

• Resource discovery.

  • Searching without a road map.

• Efficient resource use.

  • Without change in the system (efficiency).

  • As the system changes size (scalability).

    • Size changing can mean getting larger or smaller.

• Identity and authority.

  • Could mean either requiring it or prohibiting it.

Proto-Peer-to-Peer Systems

• File Transfer Protocol (FTP).

  • No central authority or control.

  • Each host can offer and download files.

  • Each host has autonomy over the files it provides.

  • Simple client-server interactions.

  • Add-on systems like archie make indexing and searching possible.

• Domain Name System (DNS).

  • A global hierarchy, but local autonomy.

  • Each DNS server handles client requests and receives zone information from other DNS servers.

  • Each DNS domain is autonomous down to individual servers.

  • Client-server interaction with caching and time-outs.

  • The hierarchy provides efficient indexing and searching.

Peer-to-Peer Systems

• Napster

  • Half-way between client-server and peer-to-peer systems.

• Gnutella

  • One of the original peer-to-peer resource-sharing systems.

• Freenet

  • Strong emphasis on autonomy and privacy.

Napster

• History.

  • 1998, Created by Shawn Fanning.

  • June 1999, napster.com opens.

  • December 1999, RIAA sues for $10,00/song transfer.

  • April 2000, Metallica sues.

  • May 2001, Napster receives $15 million in venture capital.

  • July 2000, Napster ordered to prevent copyright infringement.

  • November 2000, Napster bought by Bertelsmann.

• Napster is proprietary.

  • OpenNap is an open-source implementation of the Napster server.

Napster Principles

• Napster is a hybrid client-server peer-to-peer network.

  • Peer and resource discovery is client-server.

  • Resource retrieval is peer-to-peer.

• Each Napster peer offers a list of resources it will supply.

  • Napster requires a formal login with the server.

• Client discovery determines resource availability and retrieval cost.

Napster Operation

Napster Implementation

• Implemented on top of TCP/IP.

• Simple packet format.

  struct NapsterPacket
    short data_size
    short packet_type
    char  data[0]
  
  enum NapsterPacketType
    login_rejected  0x0000
    login_requested 0x0002
    login_accepted  0x0003
  
    whois_query     0x005b
    whois_result    0x005c
    whois_offline   0x005d
  
    // and so on.
  

Gnutella

• History.

  • Early 2000, Developed by Justin Frankel and Tom Pepper.

  • March 2000, Released by AOL (?); slashdotted; rescinded by AOL the next day.

  • 2001, Gnutella dies at the hands of its success.

  • February 2002, Morpheus relases a gnutella-based client.

• Gnutella was proprietary, but was quickly reverse engineered.

  • It's the basis for some commercial products, such as limewire.

Gnutella Principles

• Gnutella is a pure peer-to-peer network.

  • A process is a Gnutella peer.

• The Gnutella protocol is based on modified flooding.

  • A peer queries only some of its peers.

  • Queries have a lifetime.

  • But still, this isn't scalable.

• Requests leave a trail of bread crumbs; replies follow the trail back.

Gnutella Operation

• Ping requests and replies establish initial peer groups.

  • Initial contacts are based on history lists and hope.

Gnutella Implementation

• Implemented on top of TCP/IP.

  • Flooding and TCP don't go together.

  • Message and reply duplication is a problem too.

• Simple packet format.

  struct GnutellaPacket
    char  message_id[16]
    char  function_id
    char  ttl
    char  hops_taken
    short data_length
    char  data[0]
  
  enum GnutellaFunctions
    ping         0x00
    ping_reply   0x01
    client_push  0x40
    search       0x80
    search_reply 0x81
  
  

Freenet

• History.

  • June 1999, Developed and released by Ian Clarke.

  • December 1999, Clarke steps down as lead; Freenet becomes an open-source project.

  • March 2000, Beta open-source version released.

Freenet Principles

• Freenet is a pure peer-to-peer network.

• Designed to support anonymity and resist censorship.

  • Location independent storage.

• Adaptive behavior.

  • Reduce searching costs by replicating frequently requested resources.

  • Reduce transport costs by migrating resources to areas of high demand.

• Requests forwarded to most likely responders.

Freenet Operation

Freenet Implementation

• Implemented on top of TCP/IP.

• The Freenet protocol is difficult to find.

This page last modified on 17 April 2003.