% -*- TeX-master: "report.tex" -*-

% Related work  % Why our system is different from previous P2P content distribution
    % P2P
      % BitTorrent
    % CDN
      % Codeen, Coblitz
      % Coral
      % OceanStore, Pond
    % Distributed FS
      % No consistency issues in our system
      % don't want people to store data they don't need as some FS do

So far, there haven't been any distributed cache systems like
InHome. Companies interested in saving bandwidth tend to use a
centralized caching proxy server -- an effective, but costly
solution. In addition a centralized caching proxy can only work in
situations where there is a central point of entry for data into the
system. As more and more organizations use more than one ISP, a
centralized solution become more impractical.

The closest prior work to InHome are Content Distribution Networks like 
Codeen\cite{wang04:_codeen}. Although these too are distributed caching systems, 
CDNs are focused on relieving server pressure -- a specific set of content 
providers replicate their data on these networks to reduce the number of data 
requests they process. In contrast, InHome is focused on reducing the bandwidth 
of a specific set of users, storing their most commonly accessed data.

There has been similar work in reducing external Bittorrent traffic. Since
Bittorrent forms a significant chunk of cross-ISP bandwidth, systems focusing on 
reducing external Bittorrent traffic can make a significant impact on global 
bandwidth usage. Ono\cite{choffnes08:_taming_torrent} is a project from 
Northwestern that piggybacks off existing CDNs to locate local peers for 
Bittorrent downloads. Although there are no concrete measurements determining 
how much bandwidth is saved, Ono transfers draw at least 33\% peers from the 
local network on average. In addition, a study from Stanford on local peer 
selection has shown that such schemes can reduce the number of redundant copies 
of data downloaded by an ISP from 50 to 4.

Our system is more general than Ono -- with the appropriate plugin,
a Bittorrent client could discover whether a copy of the data was
located on the local network and download it if possible. Bittorrent 
was designed to download from a large number of low-throughput connections;
for local area networks, its easier to directly download the file from a single
peer.