Balasubramaneyam Maniymaran

1
Build One, Get One Free Leveraging the
Coexistence of Multiple P2P Overlay

• Balasubramaneyam Maniymaran
• Marin Bertier
• Anne-Marie Kermarrec
• ICDCS 07
• ???? ??? ??
• ????69621036 ???

2
Outline.

• Introduction.
• Motivation.
• Architecture.
• Simulation and result.
• Conclusion.

3
Introduction.

• Pastry
• Gossip
• C-gossip
• Random peer sampling(RPS)

4
Introduction.(cont)

• Pastry
• Structure P2P overlay
• Hash based
• Leaf set
• Routing table
• Pastry routing table exponentially decreases with
  the level. (from top to bottom).

5
Introduction.(cont)

• Pastry
• proximity neighbor selection (PNS).
• In PNS, a Pastry node fills its routing table
  with nodes that are closer in terms of network
  latency.
• PNS is implemented using a gossip-like protocol.
• Pastry node periodically chooses a node from its
  routing table from a random row and column and
  exchanges routing table entries belonging to the
  specific row.
• PNS significantly increases the message overhead
  of the Pastry protocol.

6
Introduction.(cont)

• Gossip protocol
• In such protocols each node periodically
  exchanges information with another node randomly
  chosen from its view so that the information is
  spread in the system in an epidemic manner.
• The gossip protocol can also be applied to create
  node clusters in application-specific overlays
  using different proximity measures.

7
Introduction.(cont)

• C-grossip
• In our model each peer maintains a dynamic list
  of semantic neighbors, called its semantic view,
  of fixed small size l.
• We are essentially seeking to pick peers Q1,Q2,
  ...,Ql for peer Ps semantic view, such that the
  sum ?li1 S(P,Qi) is maximized.
• S(FP ,FQ) FP FQ.
• In our design we decouple these two aspects by
  adopting a two-layered set of gossip protocols.

8
Introduction.(cont)

• The lower layer, called CYCLON, is responsible
  for maintaining a connected overlay and for
  periodically feeding the top-layer protocol with
  nodes uniform randomly selected from the network.
• The top-layer protocol, called VICINITY, is in
  charge of focusing on discovering peers that are
  semantically as close as possible, and of adding
  these nodes to the semantic views.

9
Introduction.(cont)

• RPS
• The aim of this service is to provide every node
  with peers to exchange information with.
• Peer selection
• Periodically, each node selects a peer to
  exchange membership information with. This
  selection is implemented by the function
  selectPeer() that returns the address of a live
  node as found in the callers current view.

10
Introduction.(cont)

• View propagation
• Once a peer has been chosen, the peers may
  exchange information in various ways.
• View selection
• Once membership information has been exchanged
  between peers and merged as explained above,
  peers may need to truncate their views in order
  to adhere to the c items limit imposed as a
  protocol parameter.

11
Motivation.

• As the number of P2P applications increases, it
  is likely that several P2P overlay networks
  cohabit on a network so that the best overlay
  could be chosen depending on the application.
• We argue that this cohabitation should be
  leveraged to create a joint overlay in which the
  overlays share information to reduce the
  maintenance cost while keeping the same level of
  performance.

12
Architecture.

• Composition of P2P overlay
• The primary component becomes the fundamental
  component of the system and has to be strongly
  maintained, while the secondary component
  sustains less aggressive maintained.
• The leaf set of the Pastry overlay is identified
  as its primary component. The routing table is
  the secondary component and is weakly constrained
• The primary component in the C-Gossip is the
  cluster view. The RPS view is the secondary
  component.

13
Architecture.(cont)

• A node may use the leaf set of the Pastry
  protocol to implement the RPS view in the
  C-gossip, while the nodes from the cluster view
  may be used to fill up the Pastry routing tables.
• This combination relieves the system of the
  maintenance of two gossip protocols the PNS
  gossip in Pastry and the RPS gossip in C-gossip.

14
Architecture.(cont)

• Design of joint overlay
• Each peer in the joint overlay maintains three
  states a leaf set and a routing table as defined
  in Pastry and a cluster view as defined in the
  C-Gossip protocol.
• Leaf set maintenance
• At each round, a node sends a message to all
  nodes in its leaf set. If a node a does not reply
  before the expiration of a timeout T, the node a
  is suspected to be failed and the leaf set is
  updated.
• Leaf sets are also updated upon the arrival of a
  new node.

15
Architecture.(cont)

• Cluster view maintenance
• The cluster view is updated using a gossip
  protocol.
• At each round a node chooses one node from the
  combination of cluster view and leaf set and
  exchanges its cluster view with that node. When a
  node receives a view, it merges the local and
  received views and selects the c closest nodes
  according to the defined proximity metric.

16
Simulation and result.

• We use the PeerSim simulator for our simulations.
  PeerSim is a Java based discrete-event simulator
  developed at University of Bologna, Italy.
• We had to restrict our simulations to a network
  size of maximum 50,000 nodes.
• The interest score, sij, between the nodes is
  defined as

17
Simulation and result.(cont)

• Static Network
• where the nodes join but never leave.

18
Simulation and result.(cont)
19
Conclusion.

• We have presented the architecture of a joint
  overlay in which the coexistence of a structured
  and an unstructured overlay.
• Maintaining part of each overlay provides the
  material to build and maintain the other part of
  each.