IBM Israel Research Seminars

Peer-to-Peer content distribution networks are currently being widely used and draw on a large fraction of the Internet bandwidth. However, these applications are not always designed to be network-friendly. Usually, these applications optimize download time by using all available bandwidth. As a result, long haul bottleneck links are becoming congested and the network load is not always balanced. We believe that truly scalable applications must take these issues into consideration.

In this talk, I will introduce the Julia content distribution network. The innovation of the Julia content distribution network is in reducing communication cost, improving the network load and reducing the usage of long haul links. Using extensive simulations of the Julia algorithm vs. state-of-the-art content distribution networks, namely BitTorrent, we examine the Julia algorithm performance. For evaluating the algorithm on the WAN, we have implemented a light weight client which was tested using more than 250 Planetlab machines. We discuss the relation between simulation results and the real world deployment. While achieving slower average finishing time relative to BitTorrent we reduce the total communication cost in the network by approximately 33%. Furthermore, our protocol achieves a better load balancing of the network resources, especially over trans-Atlantic links. Improved load balancing of nodes work creates better incentives to participate in the network. Since each node is required to contribute to the network, the same amount of data is serviced by the network.

This is a a joint work with Prof. Dahlia Malkhi (MS Research + HUJI).

About the Speaker
Danny Bickson received his M.Sc and B.Sc degrees in 2003 and 1999 respectively, from the Hebrew University of Jerusalem. He is currently a graduate student of Prof. Danny Dolev and Prof. Dahlia Malkhi of the DANSS (Distributed Algorithms Networking Security and Systems group) in HUJI. His research interests concentrates in the area of large scale Peer-to-Peer content distribution.