Danny Bickson, HUJI

1
Final Report on Content Distribution and Storage

• Danny Bickson, HUJI
• March 3, 2008
• Evergrow Final Review, Turin

2
Previous Activities

• Two open source projects
• The Julia content distribution network
• Myriad Store
• Data placement using message passing
• Multiple Message gossip

3
Talk outline

• Covering the activity in SP3m during 2007.
• The Julia content distribution network
• Support for real time streaming
• Extension for network coding
• Rating users in social networks
• Data aggregation in sensor networks
• Work reported in deliverable D3.m3

4
WP interactions
Message passing algorithms (SP4)
Applications to Peer-to-Peer networks (SP3)
Everlab (SP2)
Topologies from SP1
5
Our approach

• Using message-passing algorithms from the complex
  systems domain to be used in Peer-to-Peer
  networks
• Real time streaming BP (belief propagation)
• Aggregation in sensor networks BP
• Rating users Gaussian BP
• Algorithms are distributed, local interaction
  between neighbors, asynchronous
• Extended scope of this workpackage
• New topologies social networks and sensor
  networks
• New tools network coding, Gaussian BP
• New problems streaming media, clustering, rating
  users

6
Using network coding

• The problem
• Network coding is an emerging technique for
  increasing availability of file chunks in the
  network.
• It is not widely deployed in practice
• The solution
• We propose several simple
• heuristics that will enable easier
• implementation of network coding
• Work reported in IEEE P2P 07

7
Experimental results
8
Rating users in social networks

• The problem
• During download we need to evaluate the quality
  of data received
• We would like to rank our neighbors for
  optimizing download speed
• The solution
• Distributed rating algorithm where each node
  locally rates its neighbors
• Best neighbors are selected for download
• Work reported in IEEE P2P 07 and PPNA 08

9
Example datasets
MSN Messenger
DIMES Internet topology
Bloggers topology
10
Experimental results
11
Data Aggregation in Sensor Networks

• The problem
• We would like to aggregate observed data in
  sensor network.
• Low computation power, high failures, limited
  energy, wireless transmissions
• The solution
• Build a hierarchy of cluster heads
• Selection is done locally using BP
• Extensive simulations using TinyOS
• Paper accepted to EWSN 08

12
Experimental Results
Clustering events
Data points collected
Dropped packets
13
Publications this year

• Tal Anker and Danny Bickson and Danny Dolev and
  Bracha Hod, Efficient Clustering for Improving
  Network Performance in Wireless Sensor Networks,
  In European Conference on Wireless Sensor
  Networks (EWSN'08).
• D. Bickson and D. Dolev and Y. Weiss. Resilient
  Peer-to-Peer Streaming. PPNA Journal, submitted.
• Danny Bickson, Peer-to-Peer Rating. In the 7th
  IEEE Peer-to-Peer Computing, Galway, Ireland,
  Sept. 2007.
• Danny Bickson and Dahlia Malkhi, A unifying
  framework for rating peers and data items in
  Peer-to-Peer and social networks. In PPNA
  Journal. Accepted, January 2008.
• Danny Bickson, BitCode A BitTorrent Clone using
  Network Coding. In the 7th IEEE Peer-to-Peer
  Computing, Galway, Ireland, Sept. 2007.
• Danny Bickson (editor) Deliverable D3.m3,
  December 2007, Evergrow EU project.

14
Future work

• Extend the clustering in sensor network to
  support more advanced data aggregation algorithms
  (i.e. collaborative signal processing)
• Extend rating in social networks to include
  collaborative filtering
• Distributed monitoring using message passing
  algorithms

15
THE END

• Thank You!