Biomedical Engineering Reference
In-Depth Information
Figure 4 shows that when the number of copy
agents is increased, which is achieved by increas-
ing the threshold Tf , the reduction of the overall
entropy is lower. For example, with Tf =3, the value
of the overall entropy decreases from the initial
value of about 1 (maximum disorder) to less than
0.72. With Tf =9, however, the entropy is only re-
duced to about 0.87. As an extreme case, virtually
no entropy decrease is observed if all the agents
operate in copy ( Tf =10), which confirms that the
presence of move agents is necessary to perform
an effective descriptor reorganization.
It can be concluded that copy agents are useful
to replicate and disseminate descriptors but it is the
move agents that actually perform the descriptor
reorganization and are able to create Grid regions
specialized in specific classes of resources. A
balance between the two features (replication and
reorganization) can be performed by appropriately
tuning the pheromone threshold.
As opposed to the indices described so far,
the processing load L , i.e., the average number of
agents per second that are processed by a peer,
does not depend on the pheromone threshold, but
on the number of agents and on the frequency of
their movements across the Grid. Recalling that
the number of agents Na is approximately equal
to the number of peers Np times the mean number
of agents generated by a peer, Ngen , L can be
obtained as follows:
formed discovery protocol, in order to maximize
the number of discovered resources and minimize
the response time. Such a discovery protocol,
namely ARDIP (Ant-Based Resource Discovery
Protocol), is based on the notion of representative
peers . As descriptors of a class are accumulated
in a Grid region, the peer that, within this region,
collects the largest number of descriptors is
elected as a representative peer for this class. The
objective of a discovery operation is to direct a
query message to a representative peer as fast as
possible, since this peer likely maintains a large
number of useful descriptors.
The key features of the ARDIP protocol are
discussed in the following, while a detailed discus-
sion and evaluation can be found in (Forestiero
& al., 2007). With ARDIP, a discovery operation
is performed in two phases, one blind and one
informed . In the blind phase, the random walks
technique is adopted (Lv & al., 2002): a number
of query messages are issued by the requesting
peer and travel the Grid in a blind (random)
fashion. Whenever a query gets close enough
to a representative peer, the search becomes
informed and the query is driven towards this
representative peer. When the query arrives at
the representative peer, a queryHit message is
generated and gets back to the requesting peer
by following the same path in the opposite di-
rection. The proximity of a representative peer
is detected through another kind of pheromone
mechanism, which is inspired by the behavior of
ants that search for a food source. When a query
message first discovers a representative peer in
its random path, the queryHit message leaves
an amount of pheromone in the first peers of its
return journey, and this pheromone can be later
recognized by other peers that casually approach
this representative peer.
The semi-informed protocol aims to combine
the benefits of both blind and informed resource
discovery approaches which are currently used
in P2P networks (Tsoumakos & Roussopoulos,
2003). In fact, a pure blind approach (e.g. using
N
N
a
gen
L
=
T
N
T
mov
p
mov
With the parameter setting discussed before,
each peer receives and processes about one agent
every 120 seconds, which can be considered an
acceptable load.
RESOURCE DISCOvERy
The logical resource organization achieved by
ARMAP can be exploited through a semi-in-
Search WWH ::




Custom Search