P2P Applications - Peer-to-Peer Computing: Applications, Architecture, Protocols, and Challenges - page 8

maintaining a more or less stable level of performance (again in terms of
average download time, or in terms of availability). Scalability is there-
fore also a system-oriented performance metric.
Server Cost. As will be evident from the surveys below, a practical P2P
system still needs a good number of servers to support many useful sys-
tem functions, e.g., tracking existing peers, organizing the connections
among peers, etc. Thus, a useful indicator about the di culty in imple-
menting a P2P system is the total cost required to install these servers.
Another variable in the cost function is the bandwidth fees required for
these servers.
In the remainder of this chapter, we first briefly survey P2P applications
that are designed for computation sharing. This is followed by discussions on
P2P applications for the classical file sharing service. We then move on to
survey P2P media applications—voice and video. A summary is presented at
the end.
2.2 Distributed Processing
2.2.1 Internet Computing
One early application of a P2P computing model is to share the pro-
cessing load among many decentralized machines. The rationale is that for
many “pleasantly parallel” computing problems (i.e., large scale data parallel
problem with very little or even no dependency among parallel tasks), the
aggregate processing power of a large number of machines can match the pro-
cessing power of an expensive supercomputer. This is similar in spirit to Grid
Computing [Butler et al., 2000, Foster and Kesselman, 1999].
SETI@Home [Anderson et al., 2002,SETI@Home, 2009], launched in 1999,
is as yet the largest effort in distributed processing in terms of participants. It
is reported that it currently has over 5 million users worldwide. The objective
of SETI (Search for Extra-terrestrial Intelligence) is to exploit the aggregate
computing power of a large number of computers actively linked to the Inter-
net in order to process the daunting quantity of radio signal data gathered at
the Arecibo Observatory in Puerto Rico. The “search” is in fact a detection
problem in the sense that a large quantity of radio signal data are mined so as
to check whether there are some unusual signals which could possibly be sent
by extra-terrestrial creatures from a distant planet. A participating computer
obtains signal data from the central server at Berkeley and then processes
the data using a client program downloaded from the server site. Results are
then sent back to the server for further analysis. As a “pleasantly parallel”