Information Technology Reference
In-Depth Information
3 The Proposed Trust-Aware Algorithm
In this section, we describe the factors that are taken into account for designing the
proposed scheme, and then present the details of the scheme.
(1)
Network topology and load
: Using the approach described in [5] and [6], the
network has been modeled as a
power law graph
. In a power law network, degree
distribution of nodes follows power law distribution, i.e. fraction of nodes having
degree
L
is
L
-k
where
k
is a network dependent constant. Prior to each simulation
cycle, a fixed fraction of peers chosen randomly is marked as malicious. As the algo-
rithm proceeds, the peers adjust topology locally to connect those peers which have
better chance to provide good files and drop malicious peers from their neighborhood.
The network links are categorized into two types:
connectivity link
and
community
link
. The connectivity links are the edges of the original power law network which
provide seamless connectivity among the peers. To prevent the network from being
fragmented they are never deleted. On the other hand, community links are added
probabilistically between the peers who know each other. A community link may be
deleted when perceived trustworthiness of a peer falls in the perception of its
neighbors. A limit is put on the additional number of edges that a node can acquire to
control bandwidth usage and query processing overhead in the network.
(2)
Content distribution
: The dynamics of a P2P network are highly dependent on
the volume and variety of files each peer chooses to share. Hence a model reflecting
real-world P2P networks is required. It has been observed that the peers are in general
interested in a subset of the content on the P2P network [7]. Also, the peers are often
interested only in files from a few content categories. Among these categories, some
are more popular than others. It has been shown that Gnutella content distribution
follows
zipf
distribution
[8]. Keeping this in mind, both content categories and file
popularity within each category is modeled with
zipf distribution
with
α
= 0.8.
Content distribution model
: The content distribution model in [8] is followed for
simulation purpose. In this model, each distinct file
f
c ,r
is abstractly represented by the
tuple (
c, r
), where
c
represents the content category to which the file belongs, and
r
represents its popularity rank within a content category
c
. Let content categories be
C
= {
c
1
,
c
2
,…,
c
32
}. Each content category is characterized by its
popularity rank
. For
example, if
c
1
= 1,
c
2
= 2 and
c
3
= 3, then
c
1
is more popular than
c
2
and hence it is
more replicated than
c
2
and so on. Also there are more files in category
c
1
than
c
2
.
Table 1.
Hypothetical content distribution in peer nodes
Peers Content categories
P
1
{
C
1
, C
2
, C
3
}
P
2
{
C
2
, C
4
, C
6
, C
7
}
P
3
{
C
2
, C
4
, C
7
, C
8
}
P
4
{
C
1
, C
2
}
P
5
{
C
1
, C
5
, C
6
}
Each peer randomly chooses three to six content categories to share files and shares
more files in more popular categories. Table 1 shows an illustrative content distribu-
tion among five peers. The category
c
1
is more replicated as it is most popular. The
Search WWH ::
Custom Search