A Greener & Faster BitTorrent Architecture – WOW! With Serious Limitations
With the recent constant increase in internet related energy consumption, researchers are trying to address one main contributing factor – Running PCs 24/7 to run BitTorrents. A new architecture is being proposed to reduce the energy consumption and also reduce the download time. While promising, the architecture comes with a major limitation.
It is estimated that 74 TeraWatts hours (TWh) per year of electricity is consumed by internet. This 74 TWh is comprised of the energy consumed at data centers as well as the individual PCs. In 2007 data centers accounted for about 2 TWh per year, while office and home computers accounted for 16 TWh per year. Further, 21% of PCs used at work are never switched off, contributing to wastage of about 1.5 TWh per year. Of the variety of reasons for the computers not being turned off, P2P applications such as BitTorrent play a vital role. Recent studies report that 40 to 73% of the internet traffic is P2P and BitTorent is the most popular p2p protocol. Users are always in the quest for “which bittorrent program is the best”. Based on these statistics, the group of researchers focus on a proxy based architecture for BitTorrent.
Briefly, the original BitTorrent architecture which is an unstructured overlay network, comprises of peers (the actual end computers which upload and download different fragments of the file at the same time), tracker (the computer which helps the peers discover newer peers, enforce sharing ratio etc) and the torrent file (which lets the peers know about the tracker’s location and other basic meta data).
The proposed modified architecture makes a huge assumption, where several BitTorrent users share the same LAN network. A single computer is used for the proxy and a computer which will be left on always for other reasons is preferred. The proxy acts as a peer to the rest of the world. Internally, a simple client/server architecture is implemented where in, any individual computer (client) which needs to download a file, notifies the proxy about it. The client can then be turned off. The Energy Saving BitTorrent Module (ESBT) daemon running on the server (proxy) keeps monitoring for requests to download from the clients and translates them as a torrent download request to be initiated from the BitTorrent client installed on the server. A similar protocol is also available for uploading files. As the proxy lands up sharing more than individual clients would, the download speed is also increased. Whenever the client comes back, a status update is performed and if the download is completed, the files are transported through the LAN.
Modified BitTorrent Architecture with ESBT Proxy Server and Clients
Performance evaluation on the architecture indicates that, if the proxy power consumption is ignored, about 95% energy savings is got and if the proxy power consumption is considered, starting from two clients, the savings linearly increase. The other important performance criterion – download speed, had greatly improved (average download times decreased by 22% using a proxy versus not using a proxy). The reason for this reduction of download speed was explained by the increased number of files and/or fragments of files available for upload. Uploading a file which was very popular resulted in further reduction of download time for all the other torrents. [Though not explicitly stated, we believe that this reduction in download speed would be possible only if all/most of the downloads and uploads were from the same tracker.]
The researchers provided the statistic that 21% of users do not turn of the computers at work, but do not provide how much of them run BitTorrent (in a work environment, where a LAN would mostly exist) and also fail to point how much of home users have their computers running 24/7. Without these two statistics, it is not sure how the proposed architecture would help on a large scale. We deem that, assuming that BitTorrent users would be on a LAN is a very liberal assumption, as most users would be downloading from home networks rather than work networks.
Jounral Refernce: http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=5202284&isnumber=5202209
