Adaptive Retrieval in a P2P Cloud Datacenter - Multimedia Database Retrieval: Technology and Applications - page 210

availability (continuous service if one or more servers are unintentionally out of
service) and efficiency (access from the geographically closest server). P2P network
is a special case of such a network, where each node in the network behaves as a
database server.

8.2.2 Application of a Multimedia Retrieval System

in a P2P Datacenter

Figure 8.1 shows the application of a multimedia retrieval system in a P2P
datacenter using centralized and decentralized architectures. In this system, content
providers can be cloud providers or cloud customers. The providers transfer files
including images, videos, and plain text into datacenters from any device, such as
smart phone, tablet computer, or other lightweight device. The search starts with the
cloud customer issuing a query request. The datacenter receives the query and takes
charge of search processing and the query results are sent back to the customer. In
the cloud datacenter, when a peer has a file to share, it makes it available to the rest
of the peers. This file may be copied to some interested peers, and become available
to the group. Since lists of peers may grow and shrink, the challenge is to locate files
and keep track of the locations of the files.
In a centralized P2P network, as shown in Fig. 8.1 a, the search directory (listing
of the peers and what they offer) is centralized, but the storing and serving of files
are done using the P2P paradigm. For this reason, a centralized P2P network, such
as Napster [ 221 ] is referred to as a hybrid P2P network. In this type of network, a
peer first registers itself with a server, and sends a list of all the files to the server.
A cloud customer looking for a file goes to the server and issues a query. The server
searches its directory and responds with the IP addresses of the peers that are sharing
the file. The cloud customer then contacts one of the peers and downloads the file
directly. The search index is constantly updated as nodes join or leave the peer. This
offers greater simplicity in the maintenance of the search directory. However, using
centralized networks has some drawbacks, such as the accessing of the search index
generating a huge traffic load which slows down the system.
Figure 8.1 b, c shows a decentralized P2P network that does not depend on a
centralized directory system. In this model, peers arrange themselves into an overlay
network as either an unstructured (Fig. 8.1 b) or a structured (Fig. 8.1 c) logical
network. An unstructured P2P network has nodes that are linked randomly, whereas
a structured P2P network has nodes that are linked according to a predefined set
of rules for efficient query routing. The unstructured P2P network, such as Gnutella
[ 222 ], works by broadcasting the query with non-zero Time-To-Live (TTL) to all the
neighboring hosts. It may find a node with the desired file after a few levels of search
in the network. However, the system causes a lot of network flooding during a query
[ 223 ]. In Sects. 8.4 and 8.5 , the retrieval system adopts an unstructured P2P network
using a strategy of searching within the neighborhood community. This community