Structure Overlay Networks and Chord

1
Structure Overlay Networks and Chord

• Presentation by Todd Gardner
• Figures from Ion Stoica, Robert Morris, David
  Liben-Nowell, David R. Karger, M. Frans Kaashoek,
  Frank Dabek, Hari Balakrishnan, Chord A Scalable
  Peer-to-peer Lookup Protocol for Internet
  Applications. To Appear in IEEE/ACM Transactions
  on Networking

2
Peer-to-peer networks, History

• The first large peer-to-peer network was Napster
• As we went over in class, Napster relied on a
  central server to store all routing information,
  so it was only sort of peer-to-peer

3
Peer-to-peer networks, History

• The later peer-to-peer networks, like Gnutella,
  are fully distributed, but dont have the
  guarantees Napster had for finding files
• If a file lay beyond your number of hops on the
  network, you might not find it

4
Structure Overlay Networks

• Introducing Chord (and others)
• The idea is to build a peer to peer network, but
  to offer guarantees about finding data if it is
  on the network
• This is accomplished by enforcing a topology on
  the network

5
What do I mean by enforcing a topology?

• This is where it gets complicated
• Nodes in a Gnutella-like network join
  willy-nilly, using a reference to a node on the
  network, and having a potentially inefficient
  topology for searching

6
Attempting to correct the problem

• A simple solution would just be to publish well
  known nodes
• This is just recreating Napster

7
Chords Approach

• Whenever a node joins a Chord-type network, it
  joins a ring
• All members of the network are on the same ring
• Your position in the ring is determined by a Key
  (in most applications, this key will be your IP
  address)

8
An example ring
9
Knowledge of the ring from the Nodes perspective

• The node should know about the next node in the
  ring the nodes successor
• The node should have a table of log n (n is the
  number of total keys) other nodes around the
  ring, with exponentially increasing keys the
  nodes finger table

10
Other information

• The node stores the previous node in the ring,
  its predecessor
• The node stores values for the distributed hash
  table

11
Storing/Retrieving Data

• Data is inserted into the ring with a Key, which
  is hashed to be put onto a node in the table
• When one wants to get data, one takes the key,
  applies the hash function to find the Key of the
  node its on, and uses its knowledge of the table

12
Best shown by example
13
So, what does this get us?

• If the nodes successors are correct, it is
  guaranteed to find the correct node
• If the nodes finger table is correct, it can do
  it in logarithmic time.
• The protocol includes functions to update these
  constantly

14
Example DNS

• Russ Cox, Athicha Muthitacharoen, Robert T.
  Morris, Serving DNS using a Peer-to-Peer Lookup
  Service, In the proceedings of the First
  International Workshop on Peer-to-Peer Systems
  (IPTPS '02), March, 2002 Cambridge, MA.