Peer to Peer A Survey and comparison of peer-to-peer overlay network schemes And so on… Chulhyun Park

2 / 26 Contents  Introduction to P2P  Structured scheme  Unstructured scheme  Conclusion

3 / 26 Introduction to P2P  Peer to Peer Server-client?

4 / 26 Introduction to P2P  Overlay?

5 / 26 Introduction to P2P  Peer-to-peer overlay network Distributed systems No hierarchical organization No centralized control Overlayed on the IP Robust wide-area routing Efficient searching Storing, Selecting, Scalability, …

6 / 26 Introduction to P2P (cont’d)

7 / 26 Introduction to P2P (cont’d)  Distributed hash table Decentralized Scalability Fault tolerance

8 / 26 Introduction to P2P (cont’d)  Structured P2P overlay networks Network topology is tightly controlled Use distributed hash table as a substrate  Unstructured P2P overlay networks Each peers doesn’t know about whole network topology

9 / 26 Contents  Introduction to P2P  Structured scheme Pastry Chord P-Grid  Unstructured scheme  Conclusion

10 / 26 Structured scheme Pastry  General substrate of p2p application  Self-organizing, decentralized, scalable  128-bit node ID for each node  ┌ Log 2^b N ┐ step routing

11 / 26 Structured scheme Pastry – Node state  Routing table Entry (n,i) shares first n digits and (n+1)th digit is i  Leaf set Numerically closest of nodeID  Neighborhood set Closest in IP network

12 / 26 Structured scheme Pastry – Routing, Join, Failure  Routing Leaf set matching Longest prefix matching in routing table  Join Assumption : new peer A knows a nearby peer X Join message routed to numerically closest peer Z  Each row of routing table constructed along the path  Leaf of Z is almost same as that of A  Neighbor of X is almost same as that of A  Failure or Departure Recursive recovery

13 / 26 Structured scheme Chord  Scalable, distributed, self-organizing  Node and keys are distributed on the identifier circle  Node/Key identifier : hashing  Consistent hashing

14 / 26 Structured scheme Chord – Identifier circle / finger table  Key identifiers are stored in its successor node Lookup on successors  Finger table provides faster lookup Keep entry of 2 n th successors Add scalability A Chord identifier circle with 3 nodes (0, 1, 3) and 3 keys (1, 2, 6) A Chord finger table for previous identifier circle

15 / 26 Structured scheme Chord – join / stablization  Join Initialize finger and predecessor Update other nodes’ fingers  Stabilization A problem with inconsistent finger table may arise when concurrent join occurs Predecessor and Successor of each node will be corrected by stabilization

16 / 26 Structured scheme P-Grid  Scalable access structure for P2P apps  Perfectly decentralized  Tree structure  Every nodes is responsible for an interval in the key space There may be replica

17 / 26 Structured scheme P-Grid – access structure  Search can be started at any peer Prefix-based routing

18 / 26 Structured scheme P-Grid – construction  Construction is locally executed Nodes meet each other in many ways  Access structure, other operations, and so on Exact join operation is based on the shared prefix  Same prefix : divide coverage  Prefix relationship : extends shorter prefix  Share prefix : reference exchange

19 / 26 Contents  Introduction to P2P  Structured scheme  Unstructured scheme Napster Gnutella FastTrack  Conclusion

20 / 26 Unstructured scheme Napster  1 st generation of P2P network Purpose to share ‘Music’ files  Centralized search Single index server (no data in the server) File transfer is distributed (p2p)  Copyright?

21 / 26 Unstructured scheme Napster S S S P P PP P P Client machines (“Peers”) napster.com Servers Store their own files Store a directory, i.e., filenames with peer pointers Figure from the slides of I. Gupta

22 / 26 Unstructured scheme Gnutella  Flat topology, Decentralized  Join No specification in the protocol Pre-existing address, cached or known server, IRC, ….  Search Query flooding Flooding range : TTL  File transfer HTTP GET request

23 / 26 Unstructured scheme Gnutella  Message Randomly generated identifier Cached to drop duplicated messages  Message type Group membership (Ping and Pong) Search (Query and QueryHit) File transfer for Filewalled servant (Push)

24 / 26 Unstructured scheme FastTrack  Super-peer High bandwidth and computing power Collects meta-data Process queries  HTTP file transfer Directly from other peer Hash may be used to receive files from multiple sources

25 / 26 Conclusion  Various P2P network schemes Many improved aspects from traditional server-client model Structured – Pastry, Chord, P-grid, … Unstructured – Gnutella, Napster, … Legal issue?

26 / 26 References  Eng Keong Lua, et el., A survey and comparison of peer- to-peer overlay network schemes, IEEE COMM. Survey and tutorial on 2004  K. Aberer, P-Grid : A self-organizing access structure for P2P information systems, Proc. of 6 th CoopIS  A. Rowstron, P. Druschel, Pastry : Scalable, decentralized object location and routing for large-scale peer-to-peer systems, Proc. of 18 th ICDSP  I. stoica, et el., Chord : A scalable peer-to-peer lookup service for internet applications, SIGCOMM ’01   The Gnutella Protocol Specification v0.4  And so on…