Information Technology Reference
In-Depth Information
provide next generation capabilities for navigat-
ing new audio information spaces for extracting
statistical and content-based analysis of the audio
across the Internet.
This chapter touches on several objectives
for Intelligent Music Information Systems: Tools
and Methodologies. Firstly, it provides a new
P2P framework for MIR applications. Secondly,
it exposes a new paradigm for the way we create
massively distributed MIR applications by en-
abling users to contribute their own audio to the
network, which is an alternative approach to the
standard pull model employed currently. Lastly,
it represents a shift away from centralized control
and management of information by employing the
use of a decentralized MIR metadata database,
which is stored across the network participants as
a whole. We believe that this approach is scalable
and escapes from the bottlenecks associated with
management and configurability, not to mention
cost, of the alternative centralized approaches.
We believe the ideas presented here would be of
great interest to the readers of this topic.
This chapter is organized as follows. In the
background section, we discuss the background
and history of Grid and P2P technologies, and
also look at related research and work in the field
of MIR. The next section focuses on the Triana
framework, the current tools already in place for
audio-rate signal processing, and gives a descrip-
tion of how Triana is employing a decentralized
P2P framework to support MIR applications, as
well as our progress towards that goal. The third
section looks in detail at the DART system, ex-
plaining the work package assignment and results
retrieval protocols, detailing the various peers in
the DART system and illustrating their connectiv-
ity. The last main section then looks towards the
future and discusses how DART hopes to lead the
way when building next-generation MIR applica-
tions. Finally, we finish with a conclusion and a
summary of the DART framework.
Background
Recently, P2P networks have been broadly clas-
sified as using “unstructured” or “structured”
approaches to locate resources. Gnutella (http://
gnutella.wego.com/) and Kazaa (http://www.
kazaa.com/) are examples of unstructured P2P
networks, with hosts and resources being made
available through dynamic network overlays, and
without any global overlay planning. Distributed
Hash Table (DHT) based systems such as FreeNet
(Miller, Hong, Sandberg, & Brandon, 2002) and
Chord (http://www.pdos.lcs.mit.edu/chord/) use a
so-called “structured network” overlay of peers.
This structuring consists of a logical identifier
space (using hash keys) to which peers and re-
sources are mapped. Peers maintain a neighbor-
hood state about each other enabling application
level routing of messages through the network
based on the identifier space.
Within DART, we are more interested in the
creation of unstructured technologies that can
adapt to the types of connectivity associated with
Internet scale applications. To this end, we have
adopted the use of a middleware framework, called
P2PS (Wang 2005), which is capable of creating
super-peer networks across a range of heteroge-
neous devices. P2PS is a lightweight peer-to-peer
infrastructure based on XML discovery and com-
munication and provides a simple interface on
which to develop peer-to-peer style applications,
hiding the complexity of other architectures such
as JXTA (Brookshier, Govoni, Krishnan, & Soto,
2002) and Jini (http://www.jini.org/). Due to its
adoption of XML, P2PS is independent of any
implementation language, and through the use
of virtual pipes, it can support different transport
protocol (such as TCP/IP, Multicast, UDP). Fur-
ther, the reference implementation is in Java and
can therefore run on most operating systems.
BOINC is a software platform for distributed
computing using volunteered computer resources,
developed at U.C. Berkeley Spaces Sciences
Laboratory. The main goal of BOINC is to ad-
Search WWH ::
Custom Search