An Ontology-Based P2P Network for Semantic Search - Cloud, Grid and High Performance Computing: Emerging Applications - page 311

Table 1. The results for the bootstrapping process

link immediately for the affected connection. With
this backup scheme, we can minimize the disrup-
tion to query routing in the highly dynamic network
where peers frequently leave and join.

Processes Average Time Taken

Program Initialization 4.26 s

Semantic Clustering Mapping 0.251 ms/RDF triple

Joining Process 2.56 s

RELATED WORK

For testing the dynamic characteristic of the net-
work, we introduce a parameter: T ime-to- S tabil-
ity ( TS ) . We define the steady state of a peer as
the state in which a peer maintains live connec-
tions to the peers in its routing table. The steady
state of a peer may collapse if one of the follow-
ing events occurs:

Centralized RDF repositories and lookup systems,
such as RDFStore ([RDFStore) and Jena (Jena 2),
have been implemented to support the storing and
querying of RDF documents. These systems are
simpler to design and reasonably fast for low to
moderate number of triples. However, they have
the common limitations of centralized approaches,
such as single processing bottlenecks and single
points of failure.
Schema-based P2P networks, such as Edutella
(Nejdl, et al., 2003), are proposed to combine
P2P computing and the Semantic Web. These
systems build upon peers that use explicit schemas
to describe their contents. They use super-peer
based topologies, in which peers are organized
in hypercubes to route queries. However, cur-
rent schema-based P2P networks still have some
shortcomings: queries have to be flooded to every
node in the network, making the system difficult to
scale. Crespo, et al. (2003) proposed the concept
of Semantic Overlay Networks (SONs) in which
peers are grouped by semantic relationships of
documents they store. Each peer stores additional
information about content classification and route
queries to the appropriate SONs, increasing the
chances that matching objects will be found
quickly and reducing the search load. However,
queries still need to be flooded in each overlay

• Its short range contacts or long range contacts
leave the network or some of these peers
change their major semantic clusters (due
to their local data change).
• Its reference peer(s) leave the network or their
major semantic clusters change.

Queries routing may be affected when peers
are not in the steady state. The TS parameter is
measured from the time when the steady state of
a peer collapses until it reaches the steady state
again. We measure the TS of the affected peers
for different test cases and the results are sum-
marized in Table 2 (note that no backup links are
used in these cases).
In a highly dynamic network, peers leave and
join frequently; this may result in high relapse
rate. A high relapse rate may affect query routing
in the network. To prevent this, we use a backup
link for each type of connections. Once the steady
state collapses, a peer can switch to the backup

Table 2. Results on TS

Test Cases (without backup links) Average TS

Case 1: The short range contacts or long range contacts leaves the network or changes its
major cluster or cluster splitting/merging occurs

271 ms per connection

Case 2: Reference hosting nodes leave/change 87 ms per reference