BitTorrent has moved the state of the art forward in terms of scaling to large audiences. In this work we (Ashwin Bharambe, Venkat Padmanabhan and I) took a hard look at the mechanisms to see why it works so well, and if/when it doesn’t. The secret of it’s success seems to be the Tit-for-tat data exchange between peers, and a local-rarest-first request order for data, which prevents the infamous “last block problem.” We discovered the BitTorrent has slightly more difficulty scaling down than up: it can have difficulty when the seed bandwidth is low. We propose a smartseed strategy, a variant on the TFT scheme and a bandwidth estimation that improves the performance. The simulator code is available. Joint work with Ashwin Bharambe and Venkat Padmanabhan.
I can’t really say why I find this surprising, but for some reason I’ve held on to the notion that MS and the BitTorrent protocol would never be something you would find grouped together in regards to any sort of product and/or research.
Obviously this was just one of those silly assumptions that was grounded upon non-sense.
Cool! I’m always glad to discover things such as this, as it helps reinforce the notion that MS is *TRULY* interested in advancing the Art of Computer Science in *ALL* conceivable areas. Its simply impossible to provide support for every new idea that is ever developed, no matter how clever and/or disruptive it might be.
Haven’t had a chance to play with the above link, but when I do, I will report back what, if anything, I discover.