Evaluating a Bio-Inspired Approach for the Design of a Grid Information System - Advancing Artificial Intelligence Through Biological Process Applications - page 299

clearness, here the number of classes is set to 3,
and the peers are arranged in a 2-dimensional
mesh. The figure shows only a portion of the
Grid, though the simulation was performed on a
network with 2500 hosts.
In this figure, different symbols are associated
with the three classes. Each peer is visualized by
the symbol that corresponds to the class to which
the largest number of descriptors, maintained by
this peer, belong. Furthermore, the symbol size is
proportional to the number of descriptors of the
dominant class. Two snapshots of the network
are depicted: the first is taken when the ARMAP
process is initiated, at time 0, the second is taken
400,000 seconds later, in a quite steady situation.
This figure shows that descriptors are initially
distributed in a completely random fashion, but
subsequently they are accumulated and reorga-
nized by agents in separate regions of the Grid,
according to their class. The peers with a marked
border are representative peers that work as at-
tractors for query messages, as briefly described
in the next section.
A set of simulation runs have been performed
to evaluate the performance of the basic pick and
drop probability functions and the effectiveness
of the pheromone mechanism that drives the

mode switch of agents. These simulations were
performed with a Grid size Np equal to 2500 and
a number of classes Nc equal to 5. Figure 2 reports
the number of agents that work in copy mode,
Ncopy (also called copy agents in the following),
versus time, for different values of the pheromone
threshold Tf . When ARMAP is initiated, all the
agents (about 1250, half the number of peers) are
generated in the copy mode, but subsequently
several agents switch to move , as soon as their
pheromone value exceeds the threshold Tf . This
corresponds to the sudden drop of curves that
can be observed in Figure 2. This drop does not
occur if Tf is equal to 10 because this value can
never be reached by the pheromone (see formula
(4)); therefore, with Tf =10 all agents remain in
copy along all their lives.
After the first phase of the ARMAP process,
an equilibrium is reached because the number of
new agents which are generated by hosts (such
agents always set off in copy mode) and the number
of agents that switch from copy to move get bal-
anced. Moreover, if the pheromone threshold Tf
is increased, the average interval of time in which
an agent works in copy becomes longer, because
the pheromone level takes more time to reach this
threshold; therefore the average value of Ncopy

Figure 2. Number of agents in copy mode for different values of the pheromone threshold Tf