Updated the abstract to the ACM class.

[quix0rs-apt-p2p.git] / docs / abstract / abstract.tex

\documentclass{sig-alt-release2}

\usepackage[noadjust]{cite}
\usepackage{url}
\hyphenation{dis-tri-bu-ted}

\begin{document}

\conferenceinfo{SIGCOMM'08,} {August 17--22, 2008, Seattle, Washington, USA.} 
\CopyrightYear{2008}
\crdata{978-1-60558-175-0/08/08} 

\title{Leveraging Altruistic Peers to Reduce the Bandwidth Costs of Free Content Downloads}
\numberofauthors{2}
\author{
\alignauthor Cameron Dale\\
        \affaddr{School of Computing Science (student)}\\
        \affaddr{Simon Fraser University}\\
        \affaddr{Burnaby, British Columbia, Canada}\\
        \email{Email: camerond@cs.sfu.ca}
%\and
\alignauthor Jiangchuan Liu\\
        \affaddr{School of Computing Science (faculty)}\\
        \affaddr{Simon Fraser University}\\
        \affaddr{Burnaby, British Columbia, Canada}\\
        \email{Email: jcliu@cs.sfu.ca}
}

\maketitle

\begin{abstract}
We introduce the opportunity many free content distributors have to
reduce their bandwidth costs using a peer-to-peer download
mechanism. Our model for this peer-to-peer distribution is simple to
implement, backwards-compatible with the existing infrastructure,
and can be deployed incrementally. It makes wide use of a customized
distributed hash table (DHT) to facilitate the finding of peers and
the efficient downloading of files. We also present the example
solution we created for one of these distributors, the Debian
project, which is already in use by some users.
\end{abstract}

% A category with the (minimum) three required fields
%\category{H.4}{Information Systems Applications}{Miscellaneous}
%A category including the fourth, optional field follows...
\category{C.2.4}{Computer-Communication Networks}{Distributed Systems}[Distributed applications]
\terms{Design, Performance, Reliability}
\keywords{BitTorrent, P2P}

\section{Introduction}
\label{intro}

There are a large number of free content distributors using custom
package distribution systems over the Internet to distribute content
to their users: most Linux distributions (e.g. Debian, Red Hat,
Gentoo), Cygwin, CPAN, etc.
These distributors have developed many different
methods for this distribution, but they almost exclusively use a
client-server model to satisfy user requests. The contents' size,
popularity and number of users results in a large number of
requests, which usually requires a complicated network of mirrors to handle. Due
to the free nature of this content, many users are willing and able
to contribute upload bandwidth to this distribution, but currently
have no way to do this.

There are many peer-to-peer implementations available today, but
none is very well suited to this specific problem. Many distributors
make their content available in some form using BitTorrent \cite{COHEN03}, but it
is not ideally suited to an application such as this one. BitTorrent
is designed for large static collections of files, in which all
peers are interested in most or all of the content. But the packages
in this case are small, constantly being updated, and users are only
interested in a small subset of them. Other proposed implementations
suffer from some of the same, and other, problems and so none has
been well-used to address this problem.

We present a new peer-to-peer distribution model to meet these
demands, based on other successful peer-to-peer protocols such as
Distributed Hash Tables and BitTorrent.
The model relies on the pre-existence of cryptographic
hashes of the content, which uniquely identify it for a request from
other peers. If the peer-to-peer download fails, then the original
request to the server is used as a fallback to prevent dissatisfied
users. The peer can then share this new content with others through
the P2P system. This system would efficiently satisfy the demands of
a large number of users, and should significantly reduce the
currently substantial bandwidth costs of hosting this content.

We also present a sample implementation based on the Debian package
distribution system. This implementation is simple, makes use of
mostly well-known technologies, and so serves as an example for
other free content distributors of the opportunity that can be easily met
with such a system.

\section{Problem}

There are a large number of groups using the Internet to distribute
their free content. This content is made available from a free
download site, which usually requires a network of mirrors to
support the large number of requests. This content almost always
supports some kind of file verification using cryptographic hashes
to verify the download's accuracy or authenticity.

Most Linux distributions use a software package management system
that fetches packages to be installed from a network of mirrors.
Debian-based distributions uses the \texttt{apt} program, which
downloads Debian packages from one of many mirrors. RPM-based
distributions use \texttt{yum}, and Gentoo uses \texttt{portage},
which both operate in a similar way. Other free software
distributors also use a similar system: CPAN distributes files
containing software packages for the PERL programming language,
using SOAP RPC requests to find and download files; Cygwin provides
many of the standard Unix/Linux tools in a Windows environment,
using a package management tool that requests \texttt{tar} files from
websites; two software distribution systems exist for Mac OSX,
fink and MacPorts, that also retrieve packages in this way. Also,
some systems use direct web downloads, but with a hash verification
file also available. These
hash files usually have the same file name, but with an added
extension identifying the hash used (e.g. \texttt{.md5} for the MD5
hash). Finally, there are other programs that make use of
cryptographic hashing to identify files: e.g. Git is a version
control system in which all files and changes are identified by
their SHA1 hash.

These systems all share some commonalities: the content is avaliable
for anyone to freely download, the content is divided into distinct
units (packages), users are typically not interested in downloading
all of the content available, and hashes of the content are
available before the download is attempted. We also expect that
there are many users of these systems that are motivated by
altruism to want to help out with this distribution. This is common
in these systems due to the free nature of the content being
delivered.

% Although at first it might seem that a single all-reaching solution
% is possible for this problem, there are some differences in each
% system that require independent solutions. The systems all use
% different tools for their distribution, so any implementation would
% have to be specific to the tool it is to integrate with. In
% particular, how to obtain requests from the tool or user, and how to
% find a hash that corresponds to the file being requested, is very
% specific to each system. Also, there may be no benefit in creating a
% single large solution to integrate all these systems since, even
% though they sometimes distribute nearly identical content, it is not
% identical as it has been tailored to the system, which will change
% the hash of the files and so will make it impossible to distribute
% similar content across systems.

\section{Solution}

This situation presents a clear opportunity to use a new
peer-to-peer file sharing protocol.
A downloading peer can lookup the package hash using the protocol and, if it
is found, download the file from those peers and verify the content.
If the package hash can not be found, the peer will
fallback to downloading from the original content location, and once
complete will announce to other peers indicating that it now has
the content. The original servers or mirrors thus act as \emph{seeds} for the
P2P system without any modification to them, and users are satisfied even when
there are no peers which allows the system to be deployed incrementally.

This functionality could be directly integrated into the
package management software, although this would be
difficult as the protocol should be running at all times, whereas the
package software typically only runs until the download request is complete.
Alternatively, it
it possible to implement the P2P aspect as a caching
proxy. The proxy will get uncached requests first from the P2P system,
and then fallback to the normal request from a server.

The sparse interest in a large
number of packages undergoing constant updating is well suited to
the functionality provided by a Distributed Hash Table (DHT). DHTs
require unique keys to store and retrieve strings of data, which the
cryptographic hashes used by the package management systems are
perfect for. The stored and retrieved strings can then be pointers
to the peers that have the content package that hashes to that key,
as well as any other information needed by the protocol to facilitate
the download.

To download larger files efficiently from a number of peers,
the file needs to be broken up into pieces, each with its own hash.
This method, inspired by BitTorrent, makes it very easy to
parallelize the downloading process and maximize the download speed.
Since the package management systems only make
available a hash of the entire content, the piece hashes will have
to be found through the P2P protocol.

For files smaller than the piece size, no piece hashes are
necessary. For medium size files of only a few pieces, the piece
hashes are short enough to be stored in the DHT with the pointer to
the peer to download from. For large files of 10's of pieces, the
piece hashes are generally too long to store with the pointer to
download peers. Instead, the peers will store the piece hashes for
large files separately in the DHT using as a key the hash of the
piece hashes, and include this key in their stored value for the
hash of the file. Peers downloading these large files can then
retrieve the piece hashes using a second DHT request. For very large
files for which the piece hashes are too large to store at all in
the DHT, a request to the peer for the hash (using the same method
as file downloads) should return the piece hashes.

\section{Sample Implementation}

To demonstrate the ease and effectiveness of this solution we have
created a sample implementation called \texttt{apt-p2p} which
interacts with the \texttt{apt} tool found in most Debian-based
Linux distributions. \texttt{apt} uses SHA1 hashes to verify most
downloaded files, including the large index files that contains detailed
information such as the hashes of the individual packages. Since all
requests from \texttt{apt} are in the form of HTTP downloads from a
server, the implementation takes the form of a caching HTTP proxy.

The DHT used is based on Khashmir, which is an implementation of
Kademlia \cite{kademlia} used by most of the existing BitTorrent
clients to implement trackerless operation. It has been modified to
better support the retrieval of multiple values (peers) per key, and to
improve the response time so as to satisfy users' demands for speed.
The values stored in the DHT are all bencoded dictionaries similar
to torrent files. Peers store their download location (IP address
and TCP port), as well as the piece strings (or hashes of piece
strings) as described in the previous section. Downloading is
accomplished by simple HTTP requests for the hash of the file,
which is sent to the peers identified in a DHT lookup to
have the desired file. All peers support HTTP/1.1, both as
servers and clients, which allows for pipelining of multiple
requests to a peer, and the requesting of smaller pieces of a large
file using the Range request header.

We have chosen a piece size of 512 kB, which means that 17,515
(79\%) of the almost 23,000 Debian packages will not require piece
information as they are smaller than a single piece. There are 3054
packages (13\%) that will require 2 to 4 pieces, for which the piece
hashes are stored directly with the peer download information in the
DHT. There are 1667 packages (7\%) that will require 5 to 70 pieces,
and so use a separate lookup in the DHT for the piece hashes.
Finally, there are only 62 packages (0.3\%) that require more than
70 pieces, and so need a separate request to a peer for the
piece hashes.

We have deployed the DHT part of this implementation on PlanetLab to
test it's speed and effectiveness at delivering results. We find
that, even on the overloaded PlanetLab network, the responses to
lookups of keys take less than 10 seconds on average. Since each
package download requires a lookup before it can proceed, it may
take at least 10 seconds to process a request. However, due to
\texttt{apt} pipelining up to 10 requests at a time to our proxy,
the user will see an average response time (after a short startup)
of under a second, which should be responsive enough to satisfy most
users.

This sample implementation has also been made available to all
Debian users as an installable software package. Once it becomes
more widely used, we plan to collect more usage information on the
system through crawls and data analysis.

\bibliographystyle{abbrv}
\bibliography{./all}

\end{document}