Chord-over-Chord Overlay by g7HreN4

VIEWS: 0 PAGES: 16

									Chord-over-Chord
         Overlay
         Sudhindra Rao
    Ph.D Qualifier Exam
   Department of ECECS
Outline
   Peer-to-Peer systems
   Centralized P2P systems (hybrid)
   Unstructured P2P systems (pure)
   Structured P2P systems
   Super-peer networks
   CoCO
   Analysis and Conclusion
    Peer-To-Peer Systems
   Decentralized data and resource sharing
   All computers have equal capabilities
   The resources can include:
     Processing power

     Data

     Network bandwidth

   Applications
     Redundant storage

     Permanence

     Selection of nearby servers

     Anonymity

     Search

     Authentication

     Hierarchical naming
    Centralized Server P2P
    systems - Napster
   Used in large scale sharing of files
   Single server maintains a table of data Vs
                                                           Central Server
    node
   Features
                                                                             Peer 5
     Not self-organized

     Not scalable

     Single point of failure/attack         Peer 2
     Most popular network - mp3 sharing

   Applications:                                                           Peer 4

     Napster
                                                      Peer 3
Unstructured P2P networks -
Gnutella
   Random overlay networks                 Peer 1
   No central index
   Start with nodes that know about                         Peer 5
    peer servers and flood along the
    network
   Peers find neighbors
                                                   Peer 3
   Features:
     Scalability – Flooding limited
       by TTL                             Peer 4
                                                                     Peer 2
     Keyword search

     Cannot guarantee search

   Applications:                      Peer 6               Peer 7

     Gnutella

     Freenet
    Structured P2P networks –
    Chord, Pastry, CAN
   Based on ‘Distributed Hash Tables’
   Self-organized overlay networks
   Insertion and lookup in a bounded                             Lookup (K54)
    number of hops                                           N1
                                          K54                     N8
   Features:                             N58
     Load balancing                                                   N14
     Fault-tolerance
     Decentralization
                                         N51

     Scalability                        N48
                                                                       N21
     Availability
     Flexible naming
                                               N42
   Applications:                                    N38
                                                           N32
     Chord
     Pastry
     Tapestry
     CAN
Design and Analysis
   Chord provides fast distributed computation of a hash function,
    mapping keys to nodes responsible for them
   Assigns keys to nodes with consistent hashing
   A chord node needs only a small amount of routing information
    about other nodes
   A node resolves the hash function by communicating with other
    nodes
   With high probability, the number of nodes that must be contacted
    to find a successor is an N-node network is O(log N)
   Only O(log N) fingers need be stored
   When an Nth node joins or leaves the network, only an O(1/N)
    fraction of the keys are moved
Super-Peer Networks
   Hierarchy introduces manageability
   Super-Peer networks combine features of distributed search and
    centralized search
   Super-Peer node acts as server for subset of peers
   Inherent heterogeneity in the capability of peers on the network
   Super-Peers are assigned based on processing power, network
    bandwidth, degree etc.
   Super-peers communicate by flooding to other super-peers
   Super-peer to peer communication – centralized server system
Super-peer network
   Thumb rules for design
     Increasing cluster size reduces
      aggregate load
     Super-peer redundancy makes
      system resilient
     Super-peers should have higher
      out-degree
     Minimize TTL on floods
   Drawbacks
     Flooding does not guarantee
      search success
     Super-peers can be burdened
     Flooding traffic and duplicates
Self-similarities in Mandelbrot
Set
Chord-over-Chord
Overlay(CoCO)
   Chord used in local clusters –                         Central Server
                                                           0 and 640
    Super-peer as manager
   Super-peer redundancy - by                 Super Peer              Super Peer
    assigning super-peers at the edge          1 - 127                 512 - 639


   Super-peers form a Chord overlay
    network                                    Super Peer
                                                                        Super Peer
   Super-peers maintain finger tables         128 - 255
                                                                        384 - 511

    for cluster as well as the super-peer                  Super Peer
    overlay                                                256 - 383


   Central Server consulted only if all
    Chord searches fail on the overlay
                                            Chord-over-Chord Overlay
    Chord-over-Centralized server
    Overlay (CoCO)
   Super-peers maintain a direct link to
    the Central Server
   Central Server consulted in case of                                Super Peer
                                                      Super Peer
    failed searches in local clusters
   Central Server may be single point of                     Central Server

    failure
                                                     Super Peer
                                                                         Super Peer


                                                               Super Peer




                                            Chord-over-Centralized Server Overlay
CoCO Analysis
   Number of nodes to be contacted in the local cluster of size N/m -
    O(log N/m)
   Cost of searching on Super-peer overlay - O(log m)
   Only O(log N/m) fingers need to be stored in peers and O(log m)
    additional fingers on super-peers
   When an node joins or leaves the network, only an O(m/N) fraction of
    the keys are moved and when Super peer leaves a network chord flip
    reassigns O(log N/m) + O(log m) fingers.
Discussion
   CoCO
     Uses DHT on all layers – hence resilient to failures, attacks.

     Increasing hierarchy improves manageability like Internet

     Efficient and guaranteed search results

     Joins/Leaves handled efficiently

     Super-Peer reassignment is integral part of the protocol

   Super-Peer networks using Gnutella
     Flooding can reduce efficiency

     Techniques to reduce flooding directly affect the network efficiency

     Super-peer failures are not accounted for

     Flooding on super-peers does not guarantee search results
Conclusion
    Possible applications of CoCO
      University wide P2P networks

         Each department has its own super-peer

      Company wide P2P networks

         Geographically distant networks controlled by
           administrators – super-peer assignment
      ISP controlled Napster like central server

         Strategically placed Super-peers – like Akamai caches

    Better control over the network dynamics and easy to implement
    Structured network is key to simpler administration
Thank you!

								
To top