Peer-to-peer networks constitute a widely used, cost-effective and scalable technology to distribute bandwidth-intensive content. The technology forms a great platform to build distributed cloud storage without the need of a central provider. However, the majority of todays peer-to-peer systems require complex algorithms to schedule what parts of obtained content to forward to other peers. Random Linear Network Coding can greatly simplify these algorithm by removing the need for coordination between the distributing nodes.In this paper we propose and evaluate the structure of the BRONCO peer-to-peer system, which applies random linear network coding. We focus on an experimental evaluation of the performance on 36 real nodes. The evalution shows that BRONCO outperforms regular HTTP transfers, and, with a extremely simple protocol structure, performs equivalently to bittorrent distribution. Furthermore, we evaluate the performance of different parameters and suggest a suitable tradeoff between CPU utilization and network overhead. Within the limitations of the used test environment, we have shown that networkc coding is usable in peerassisted content distribution and we suggest further improvements to reduce redundancy overhead.

Dr. Bullimore is a consultant to Alcon Surgical, Inc., Carl Zeiss Meditec AG, Digital Vision Systems, Essilor, Innovega, Inc., and Paragon Vision Sciences, Inc. Dr. Spooner is a consultant to Alcon Surgical, Inc., Carl Zeiss Meditec AG, Digital Vision Systems, Thru-Focus Optics LLC, and i2eyediagnostics, Ltd. Dr. Dishler is a consultant to Carl Zeiss Meditec AG and Revision Optics, Inc. Dr. Sluyterman is an employee of Carl Zeiss Meditec AG.

No abstract