Database Reference
In-Depth Information
cluster identification. This indexing stage facilitates online search by pinpointing
the relevant nodes without traversing all the participating nodes. Subsequently,
improvement of retrieval can be accomplished using relevance feedback within the
relevant nodes.
As an alternative to the structured P2P network, Sect.
8.4
presents an unstructured
P2P network. This is realized by the discovery of a community of neighbors
and performing automatic retrieval within nodes of the community. The search
is done by an incremental process of P2P retrieval whereby the relevant node
performs pseudo-RF and forwards the modified query to its neighbors. The system
continuously increases retrieval accuracy without transferring training samples over
the network during adaptive searching.
Section
8.5
presents pseudo-RF for video retrieval, by firstly discovering the
neighborhood community, followed by re-ranking videos via a three-layer cosine
network. The retrieval process is adaptive via intra- and inter-peer signal propaga-
tion, which can achieve high retrieval accuracy while minimizing network resources.
8.2
Distributed Database System
8.2.1
Cloud Datacenter
A cloud datacenter enhances capabilities in distributed storage and retrieval [
227
-
230
]. In order to establish high-performance datacenters for searching large volumes
of multimedia data, an appropriate topology must be chosen. The datacenter
architecture can be categorized into two types: centralized and distributed.
A centralized system, as offered by most commercial cloud services, maintains
central nodes to handle the query requests. Upon retrieving the relevant multimedia
objects according to the feature similarity measures, the universal content locator
(URL) will be returned to the requesting host. The actual content will be transferred
directly from the content server to the requesting host. The centralized systems keep
the entire feature descriptor database in centralized servers. The real multimedia
content may or may not be located on the same server. The centralized system
retrieves relevant content based on the feature-descriptor database. The drawback
of a centralized system is its limited scalability for handling growing volumes of
retrieval requests and larger multimedia databases.
In order to provide better scalability and adaptability, decentralized systems are
designed according to the Peer-to-Peer (P2P) paradigm. Each node in the P2P based
datacenter acts both as a client for requesting multimedia objects and a server for
re-distributing the multimedia objects. Since a peer can join and leave the network
at any time, a challenge of using of such a distributed retrieval system is to address
the non-guaranteed level of service of the P2P network. To localize the search, the
query packet is always associated with certain Time-to-Live (TTL) levels. Database
storage on distributed servers has been utilized in the industry to provide high
Search WWH ::
Custom Search