Scalable Overlay Network for PeertoPeer File Sharing

1
Scalable Overlay Network for Peer-to-Peer File
Sharing

• Park, Chanmo
• Networked Media Lab.
• Kwang-Ju Institute of Science and Technology
  (KJIST)
• KOREA
• cmpark_at_netmedia.kjist.ac.kr

2
Contents

• Overview
• Peer-to-Peer Computing and Core Operations
• Challenges in P2P File Sharing
• Structured Overlay Network Support for P2P
• Good and Bad
• Existing Overlay Network Approaches
• Initial Thoughts on Scalable Overlay Network
• Conclusions

3
Peer-to-Peer Computing of the Internet2

• create and administer collaboration areas
• access to the freshest data

• help businesses with large-scale
• computer processing needs using
• idle CPU cycle and disk space

• move data closer to the point
• at which it is actually consumed
• act as a network caching mechanism

• allow computing networks to dynamically work
  together
• using intelligent agents.

4
Core Operations of P2P

• Identity Operation
• Refers to the name and credentials that identify
  an entity
• Discovery Operation
• Find out peers, services, or resources
• Authentication Operation
• Verify identity (person or machine)
• Authorization Operation
• Allow the verified entity permission for specific
  actions and/or access
• Function Operation
• Application specific actions

5
Challenges in Scalable P2P File Sharing
Flooding
2
4
3
1
Loop
Loop
2
3
4
Loop
Loop
Loop
Loop
Loop
2
3
4
3

• Flooding Loop Path In Discovery Operation
• results in many duplicated packet.
• Increase the network traffics
• Reducing large number of Discovery operations
• A new topology such as structured overlay network
  should be considered

6
Structured Overlay Network Support for P2PGood
and Bad

• Good
• may reduce duplicated packets
• may remove loop in message forwarding
• Bad
• should maintain redundant link in overlay network
• Network partitioning

7
Existing Overlay Network Approaches

• AMRoute.
• ScatterCast
• End System Multicast
• ALMI

1998
2002
2001
2000
1999
1998.8 AMRoute
2001.3 ALMI
1999.7 Scattercast
1989 IP Multicast
2001.7 End System Multicast
8
AMRoute

• Goal Robust IP Multicast in mobile ad hoc
  network
• Concept
• Dynamic logical cores
• Selected by core resolution algorithm
• creates the mesh of a group
• creates the user-multicast tree
• User-multicast trees
• create a bi-directional, share tree for data
  distribution with group members
• Low scalability
• Electing a Logical core

9
ScatterCast

• Goal partitioning a heterogeneous set of session
  participants into disjoint data groups
• Concept
• SCX(ScatterCast proXy)s
• organize themselves into an overlay network of
  unicast connections (mesh)
• build data distribution tree on the top of this
  overlay structure (tree)
• using native IP Multicast

The Internet
SCX
Unicast Connection
Client
Client
SCX
Multicast Group
Client
Client
Client
10
End System Multicast

• Goal overlay network efficiency
• Concept
• construct and maintain the mesh using distributed
  algorithm
• Improving mesh quality by probing each others
• construct tree based on distance vector, latency
• Not scalable due to probing each nodes

11
ALMI (An Application Level Multicast
Infrastructure)

• Goal support of multicast groups of small size
• Concept
• Node of a session are connected via a virtual
  multicast tree
• Tree is formed as a minimum spanning tree by
  session controller

Session Controller

• handle member registration
• maintain multicast tree

RRT monitoring
Parent
Virtual Multicast Tree
Child
12
From Existing Approaches

• Mesh-first Topology
• AMRoute, ScatterCast, End System Multicast
• is suitable for multimedia multicast
• Data delivery path is constructed over mesh
• If no RPs
• Not scalable
• Not suitable for P2P File Sharing
• Existing Approaches finally construct only one
  data delivery path
• Mesh topology is proper for failure recovery
• In the case of P2P File Sharing, Network
  efficiency is not issue
• For Scalability, Tree-first can be considered as
  possible as can

13
Initial Thoughts on Scalable Overlay Network for
Peer-to-Peer File Sharing

• Challenges are reducing duplications and
  preventing loop
• Node Grouping (Tree)
• For reducing duplications
• Flooding Messages are forced to sent to limited
  scoped nodes by grouping
• Shift from Undetermined path to Determined path
• For Loop Avoidance
• Tree is constructed at first and maintained

14
Initial Thoughts on Scalable Overlay Network for
Peer-to-Peer File Sharing(2)

• we construct tree for grouping nodes
• tree
• Is a scope of message flooding
• Is a shared tree
• Is a data path along which all message are sent
• Single Data Path within a tree
• Loop avoidance
• Messages are sent to parent and child nodes
  except of received node.

15
Initial Thoughts on Scalable Overlay Network for
Peer-to-Peer File Sharing(3)

• In case of Exchanging messages between trees
• New connection type (such as a InterTreeConnection
  ) is introduced to distinguish link within a tree
  from link between trees
• These types of connections are determined when a
  new tree is created
• New tree is created when limited number of nodes
  is reached
• These type messages are used when a node
  connected to other tree sends messages.

16
Conclusion

• We show Problems of Unstructured P2P
• Flooding
• Looping
• Overview the existing Overlay Network
• Our Initial Thoughts on Scalable Structured
  Overlay Network
• Grouping into a tree for reducing duplicated
  messages
• a shared data path for loop avoidance
• Future Works
• Consider proper group size and Recovery mechanism
  from failure
• Implement our thoughts

17
Thanks!