The success of SETI@home and similar projects led to the development of the BOINC platform, 3 which has been used to develop many cycle-sharing P2P systems in use today. At the time of this writing, BOINC has more than half a million active peers computing on average 5. 42 petaFLOPS (floating-point operations per second). For comparison, a modern PC performs on the order of a few tens of GFLOPS (about five orders of magnitude fewer), and the world’s fastest su-percomputer as of August 2010 has a performance of about 1. 76 petaFLOPS.

Other applications. Other types of P2P applications have seen significant use, at least temporarily, but have not reached the same levels of adoption as the systems we describe here. Among them are applications that leverage peer-contributed disk space to provide distributed storage. Freenet9 aims to combine distributed storage with content distribution, censorship resistance, and anonymity. It is still active, but the properties of the system make it difficult to estimate its actual use. MojoNation36 was a subsequent project for building a reliable P2P storage system, but it was shut down after proving unable to ensure the availability of data due to unstable membership and other problems.

P2P Web content distribution networks (CDNs) such as CoralCDN16 and CoDeeN35 were deployed as research prototypes but gained widespread use. In these systems, a set of cooperating users form a network of Web caches and name servers that replicates Web content as users access it, thereby reducing the load on servers hosting popular content. During its peak usage, CoralCDN received up to 25 million hits per day from one million unique IP addresses.

Many more P2P systems have been designed and prototyped, but either were not deployed publicly or had small deployments. Examples include systems for distributed data monitoring, management and mining, 26, 37 massively distributed query processing, 19 cooperative backup, 11 bibliographic databases, 33 serverless email, 24 and archival storage. 23

Technology developed for P2P applications has also been incorporated into other types of systems. For in-

While the earliest
and most visible
P2P systems were
mainly file-sharing
applications,
current uses of
P2P technology are
much more diverse
and include the
distribution of data,
software, media
content, as well as
internet telephony
and scientific
computing.

stance, Dynamo, 13 a storage substrate that Amazon uses internally for many of its services and applications, uses distributed hash tables (DHTs), which we will explain later. Akamai’s NetSessiona client uses P2P downloads to increase performance and reduce the cost of delivering streaming content. Even though these systems are controlled by a single organization and thus do not strictly satisfy our definition of a P2P system, they are based on P2P technology.

While P2P systems are a recent invention, technical predecessors of P2P systems have existed for a long time. Early examples include the NNTP and SMTP news and mail distribution systems, and the Internet routing system. Like P2P systems, these are mostly decentralized systems that rely on resource contributions from their participants. However, the peers in these systems are organizations and the protocols are not self-organizing.

While the earliest and most visible P2P systems were mainly file-sharing applications, current uses of P2P technology are much more diverse and include the distribution of data, software, media content, as well as Internet telephony and scientific computing. Moreover, an increasing number of commercial services and products rely on P2P technology.

how Do P2P Systems Work?

Here, we sketch some of the most important techniques that make P2P systems work. We discuss fundamental architectural choices like the degree of centralization and the structure of the overlay network. As you will see, one of the key challenges is to build an overlay with a routing capability that works well in the presence of a high membership turnover (usually referred to as churn), which is typical of deployed P2P system. 28 We then present solutions to specific problems addressed in the context of P2P systems: application state maintenance, application-level node coordination, and content distribution.

Note that our intention in this presentation is to provide representative a See Akamai NetSession Interface Overview at http://www.akamai.com/html/misc/akamai_ client/ netsession_interface.html/.