Achieving High Bandwidth Data
Transfers with The Right Incentives


Related Work

Under the Intelligent Nation 2015 master-plan, IDA will be building an ultra high-speed National Broadband Network and a pervasive Wireless Broadband Network. In this proposal, we describe a new project to exploit the latest peer-to-peer technologies to develop a new class of high-throughput data transfer algorithms that is applicable to important applications such as (i) the online distribution of software and media files, and (ii) the online distribution of real-time media.

Peer-to-peer architectures were first proposed for managing flash crowds and improving data transfer. The first truly ubiquitous and successful peer-to-peer file sharing algorithm is BitTorrent. Unfortunately, BitTorrent has mainly been used by online pirates to disseminate illegal content and BitTorrent is not currently optimized for high speed data transfer.  The known drawbacks with BitTorrent include the following: (i) clients typically leave the system once their downloads are complete and this degrades the download performance significantly for the nodes that join a torrent late or have lower bandwidths; (ii) there are insufficient incentives for nodes to upload data to other nodes at maximal capacity and so it is unlikely that BitTorrent will be able to utilize all the available inter-nodal bandwidths efficiently; and (iii) high bandwidth nodes that join a torrent late will suffer from slow downloads despite their available bandwidth.

In this project, we will develop, implement and deploy a new peer-to-peer data transfer algorithm between a server and a number of clients, called the “Tit-for-Tat” Transfer Protocol (TFTTP), that attempts to achieve the correct incentive structures while fully exploiting inter-nodal bandwidths when downloading huge (typically > 500Mb) data files. Our protocol is incentive-compatible and is expected to achieve optimal download bandwidths for a heterogeneous set of clients, as well as cope effectively with flash crowds. In addition to this major research focus, this project will study and explore a number of fundamental research problems in peer-to-peer distributed systems.

Our approach is based on the following insight: “Given k clients that each want to download the same file, we can divide the file into k portions and give each client one portion. The clients can then exchange the portions among themselves in such a way that they are able to each re-construct the file”. While this idea is conceptually simple, implementing a practical system that attempts to optimize download speeds is extremely challenging because practical peer-to-peer systems are highly dynamic and the available download bandwidths between hosts are highly variable.


$Date: 2008/01/01 06:35:12 $