Utilizing Network Coding for File Dissemination in Peer-to-Peer Systems

Network coding was first introduced in information theory for increasing the multicast rate in the networks with directed links. We can model many of the content distribution networks and overlay networks with such a network. Therefore, the solutions introduced in information theory can be used in these networks. In this paper, we analyze using of network coding in uncoordinated cooperative content distribution systems. Reaching the optimum performance in an uncoordinated system needs blocks of the system to be "equally important". In that case, we can retrieve the whole data without need of central controller or complex algorithms. In part of this paper, we exactly define the concept of "equally important". We also introduce a new method of block generation which keeps the blocks of network "equally important" as the network evolves. Our method is decentralized, and upgrade the performance by first letting the intermediate nodes to perform coding and second making them give priority to parallel downloading. We perform several simulations to compare our method with Avalanche, a well-known peer-to-peer system using network coding. Our simulation results show that our method performs 6-10 percent better than Avalanche in several scenarios.