PeertoPeer Technologies

1
Peer-to-Peer Technologies

• Supercomputing Lab.
• Lee Hwang Jik

2
Contents

• Introduction
• A brief History
• Beyond Client-Server Model
• Deploying Internet Services by Overlaying
• Napster In Napsters wake
• Applications
• Distributed Computation
• Properties and Issues
• Future Directions

3
Introduction

• P2p
• Embodies the global sharing of resources for a
  specific task
• Homogeneous resources within a mutually-distrustfu
  l environment
• Often created for specialist uses (e.g. file
  sharing, number factorization, searching for
  alien life)
• Grid
• Flexible, high-performance, high-availability
  access to all significant resources
• Heterogeneous, specialized resources within a
  trusted environment
• On-demand creation of general-purpose powerful,
  virtual computing systems
• Computing as a utility rather than a commodity

4
A brief History

• Beyond Client-Server Model (1/2)
• Server runs for long periods of time and delivers
  storage and computational resources
• Solutions as replication, load balancing and
  request routing
• Include the clients resources in the system
• A peer consumes its own resources on behalf of
  others (a server)
• While asking other peers to do the same for its
  own benefit (a client)

5
A brief History

• Beyond Client-Server Model (2/2)
• Claims for P2p system
• No distinguished node
• Highly fault tolerant Good performance
• Scaling properties
• Ex) Pastry, Chord, CAN
• Intermediaries
• Classical distributed systems community
• Xerox Network Systems name service
• Net News (NNTP is not client-server)
• Domain Name System (zone transfers)

6
A brief History

• Deploying Internet Services by Overlaying
• Overlay systems
• Developing new infrastructures using services of
  and layering on the existing infrastructure
• Grid systems
• Use overlaying to provide unified interfaces to
  all aspects of service management
• P2p systems
• Focused on the use of overlaying to provide an
  abstraction for addressing between peer spread
  throughout the internet

7
A brief History

• Napster
• An example of a centralized p2p system
• killer application

bottleneck
8
A brief History

• Napsters wake (1/2)
• 2nd Generation Full decentralization
• Gnutella broadcast search queries between a
  peers neighbors
• Excessive network load, limiting the size of the
  network, the amount of a clients bandwidth left
• Kazaa Gnutella evolutions
• Super-peers Caching and serving common queries
  or content

9
A brief History

• Napsters wake (2/2)
• 3rd Generation Efficient Routing Substrates
• Need to a reliable way of partitioning workload
  and addressing the node responsible
• A fraction of the global state on each
• Only a low communication overhead
• Distributed message-passing object or key
  location service
• Distributed Hash Table (DHT)
• A DHT performs the functions of a hash table.
• Store a key and value pair
• Look up a value if you have the key

10
Applications

• Distributed Computation
• Be Limited
• The granularity of computations is small
• Nodes are unreliable
• External code and data distributions is hampered
  by poor latency and bandwidth
• Philanthropic computing ex) SETI_at_Home
• Genome searching, Astrogrid and code cracking
• More tightly coupled computations (Chap.10)
• More tightly coupled applications (Chap. 11)

11
Properties and Issues (1/2)

• Harnessing Resources
• DHT can hinder the ability to recognize the
  heterogeneity of resources
• Ex) CFS (Cooperative File System) weaken about
  failure
• Grid systems tend to comprise fewer, more varied,
  more specialized resources (described,
  published.. Based on own description)
• User Connectivity
• ISP uses NAT and no recognized IP Address,
  firewall
• Grids connected by well-administratered, reliable
  networks
• Collaboration and Trust

12
Properties and Issues (2/2)

• Scalability DHT - log(n)
• Proximity
• Load Balancing
• Availability
• Anonymity and Censorship Resistance

13
Future Directions (1/2)

• Sharing Computation
• Napster file sharing vs SETI_at_Home
• Be successful vs the tip of the iceberg
• Two requirement
• Infrastructure needs to handle tightly coupled
  distributed computation
• Algorithms should be designed to exploit P2p
  properties
• Auditing and Accounting
• P2p systems make accounting difficult in general
• Accounting in a Grid env. is easier
• GSAX (Grid Services Accounting Extensions)
• OGSA

14
Future Directions (2/2)

• P2p systems and Ad Hoc Wireless Network Duality
• High overheads for topology maintenance and
  message routing prohibit the use of such systems
  on wireless ad hoc networks
• load aware
• Distribute load information
• Implement a distributed congestion control scheme
• To handle mobility - topographical metric
• Anycast routing scheme
• Reducing message-passing overhead