Information Technology Reference
In-Depth Information
approaches
(counter-based, distance-based, lo-
cation-based) and (b)
deterministic approaches
(global, quasi-global, quasi-local, local). The
former methods do not guarantee full coverage
of the network, whereas the latter does provide
coverage guarantee, and thus are preferable.
The
deterministic approaches
provide full
coverage of the network for a broadcast operation,
by selecting only a subset of nodes to forward the
broadcast packet (
forward nodes
), and the remain-
ing nodes are adjacent to the nodes that forward
the packet. The selection of nodes is done by ex-
ploiting the “state” information, that is, network
topology and broadcast state (e.g., next selected
node to forward the packet, recently visited nodes
and their neighbor sets). All the categories of the
deterministic algorithms, apart from the
local
algorithms
, require (full or partial) global state
information, thus being impractical. The local
or
neighbor-designating
algorithms maintain
some local state information, that is, 1-hop
neighborhood information by periodic exchange
of “HELLO” messages, which is feasible and not
costly. In the neighbor-designating methods, the
forwarding status of each node is determined by
its neighbors. As a matter of fact, the source node
selects a subset of its 1-hop neighbors as forward
nodes to cover its 2-hop neighbors. This forward
node list is piggybacked in the broadcast packet.
Each forward node in turn designates its own
forward node list.
Remotely related to the topic of this chapter
is the issue of multicasting streaming media
(audio/video) to MANETs (Dutta, Chennikara,
Chen, Altintas, & Schulzrinne, 2003) or uni-
casting audio to 3G UMTS devices (Roccetti,
Salomoni, Ghini, & Ferretti, 2005). These works
though assume the existence of a central server
(supplier), which provisions the mobile clients
with multimedia data.
cBmIr In WIred p2p netWorks
The field of combined CBMIR and P2P net-
works is definitely very young as the inaugural
research paper dates back to 2002 (Wang, Li, &
Shi, 2002).
In this first attempt, the authors of Wang et al.
(2002) present four P2P models for CBMIR (Sec-
tions “Generalised”, “PsPs Model”, & “Hybrid”).
The four models include all centralized, decen-
tralized, and hybrid categories. Another research
based on a hybrid configuration is presented in
Tzanetakis, Gao, and Steenkiste (2004) (Section
“Decentralized Structured”). Therein the authors
propose a scalable and load balanced P2P system
with an underlying DHT-based system. C. Yang
(2003) proposed the utilization of the feature
selection and extraction process that is described
in Yang (2002) for CBMIR in a decentralized un-
structured P2P system (Section “MACSIS-P2P”).
Finally, Karydis, Nanopoulos, and Manolopoulos
(2005) (Section “Sampling Framework”) focus
on similarity searching for similar acoustic data
over unstructured decentralized P2P networks,
proposing a framework, which takes advantage
of unstructured P2P networks and minimises the
required traffic for all operations with the use of
a sampling scheme.
The following sections contain an overall de-
scription of the systems that have been proposed
for the purposes of CBMIR in P2P networks,
classified by their degree of centralization and
structure.
centralized
Wang et al. (2002) introduced CBMIR in peer-
to-peer environments by developing four systems
based on different P2P models. Two of these
models, PsCM & PsC+M, have a centralized
architecture.
Search WWH ::
Custom Search