Information Technology Reference
In-Depth Information
In what follows this chapter, we summarize
existing work on music content-based information
retrieval in peer-to-peer networks. First, neces-
sary background information on P2P networks
such as classification and searching methods are
provided in Section “Networks”. Next, in Section
“CBMIR in Wired P2P Networks” we examine
the task of CBMIR in wired P2P networks, which
has attracted significant attention in research
related to CBMIR. A number of methods as well
as systems that have been proposed are presented
therein, classified according to their degree of
centralization and structure. In the section to
follow, we study methods for CBMIR in wire-
less ad-hoc networks and present the challenges
this new field introduces. In the final section, we
conclude this chapter and present the perspective
of P2P CBMIR.
tions in
centralized
,
hybrid,
and
decentralized
.
Centralized networks maintain a central directory
in a single location (Napster), hybrid networks
maintain more than one directories in super-peers
(Kazaa) while for the decentralized (Chord) no
central directory is kept. P2P networks can also
be classified into hierarchical and nonhierarchi-
cal based on whether their overlay structure is a
hierarchy or not. It is common for decentralized
systems to have no hierarchy, while hybrid and
most centralized systems ordinarily incorporate
some degree of hierarchy. Hierarchical systems
provide increased scalability, ease in exploiting
peer heterogeneity and high routing efficiency. On
the other hand systems with no hierarchy offer
load-balance and increased resilience.
searching methods in unstructured
p2p networks
p2p netWorks
In unstructured P2P systems, no rule exists that
strictly defines where data is stored and which
nodes are neighbors of each other. Many alterna-
tive schemes (Li & Wu, 2004 offer a survey of
searching techniques in P2P networks) have been
proposed to address the problems of the original
flooding search. These works include iterative
deepening, k-walker random walk, modified
random Breadth First Search (BFS), two-level
k-walker random walk, directed BFS, intelligent
search, local indices based search, routing indices
based search, attenuated bloom filter based search,
adaptive probabilistic search and dominating
set based search. Following are some prominent
searching methods for decentralized unstructured
P2P networks.
Early works such as the original Gnutella
used flooding algorithms, which is the BFS of
the overlay network graph with depth limit D. In
the BFS, a query peer
Q
propagates the query
q
to all its neighbor peers. Each peer
P
receiving
the
q
initially searches its local repository for any
documents matching
q
and then passes on
q
to
all its neighbors. In case, a
P
has a match in its
This section of the chapter offers a concise presen-
tation of the common classification of P2P systems
based on their attributes, as well as some of the
prominent searching methods for both wired and
wireless such systems.
P2P Classification
P2P networks can be classified based on the
control over data location and network topology
in
unstructured
,
loosely structured
and
highly
structured
(Li & Wu, 2004). Unstructured P2P
networks follow no rule in where data is stored
while the network topology is arbitrary (Gnutella).
Loosely structured P2P networks have both data
location and network architecture nonprecisely
determined (Freenet). Finally, in highly structured
networks data storage and network topology are
explicitly defined (Chord). What is more, P2P
networks can also be classified according to the
number of central directories of document loca-
Search WWH ::
Custom Search