A simulation study of block management in BitTorrent

Among the existing P2P systems for content distribution, BitTorrent (BT) is the most popular one which has attracted keen attentions from both industrial and academic forces in recent years. Its superior performance is due to the multipart downloading scheme by dividing the large file into thousands of small blocks to enable the cooperative downloading among participants. Since transmissions are provoked by interested blocks only, the block distribution will seriously affects the performance of the system, i.e., robustness and throughput. As a result, how to manage the circulation of blocks is important both theoretically and practically. BT leverages on the Local Rarest First scheme to pursue the even distribution of blocks to help peers locate what they need easily. Surprisingly, how good is its performance with heterogenous networks has never received research attention before and this motivates our work. In this study, we carried out simulations to investigate the evolution of block distribution in BT. We find that the block distribution is far from optimal in terms of block frequency (with some blocks dominating the network and some becoming extinct nearly) and topology (with same blocks tending to conglomerate). We also propose a simple source coding mechanism to achieve a BT like network with much improved performance in this view.