Evaluating a Bio-Inspired Approach for the Design of a Grid Information System - Advancing Artificial Intelligence Through Biological Process Applications - page 298

Simulation runs have been performed with the
following setting of network and protocol param-
eters. The number of peers Np , or Grid size, is set
to 2500. The average connection time of a specific
peer, Plifetime , is generated according to a Gamma
distribution function, with an average value set
to 100,000 seconds. The use of the Gamma func-
tion assures that the Grid contains very dynamic
hosts, that frequently disconnect and rejoin the
network, as well as much more stable hosts. Every
time a peer disconnects from the Grid, it loses
all the remote descriptors previously deposited
by agents, thus contributing to the removal of
obsolete information. The average number of
Grid resources owned and published by a single
peer is set to 15. Grid resources are classified in
a number of classes Nc , which is equal to 5. The
mean number of agents that travel the Grid is set
to Np /2: this is accomplished, as explained in the
previous section, by setting the mean number
of agents generated by a peer, Ngen , to 0.5. The
average time Tmov between two successive agent
movements (i.e., between two successive evalu-
ations of pick and drop functions) is 60 s. The
maximum number of P2P hops that are performed
within a single agent movement, Hmax , is set to
3. The visibility radius Rv , defined in the previous
section and used for the evaluation of pick and

drop functions, is set to 1. Finally, the pheromone
threshold Tf ranges from 3 to 10.
It is worth noting that the So-Grid replication
algorithm is very robust with respect to varia-
tions of the above mentioned parameters. The
values of algorithm parameters, e.g., the number
of resources per peer, the number of resource
classes, the average connection time of a peer,
etc., can affect the rapidity of the process, and
in some cases can slightly influence the steady
values of performance indices, but the qualitative
behavior is always preserved. This can be verified
by performing “parameter sweep” simulation
through the SO-Grid Portal, as discussed in the
next section.
The following performance indices are used.
The overall entropy E , discussed in the precious
section, is used to estimate the effectiveness of
the ARMAP protocol in the reorganization of
descriptors. The Nrpr index is defined as the
mean number of descriptors that are generated
for each resource. Since new descriptors are only
generated by ARMAP agents that work in the
copy mode, the number of such agents, Ncopy , is
another interesting performance index.
Prior to the numerical analysis, a graphical
description of the behavior of the replication
algorithm is given in Figure 1. For the sake of

Figure 1. Accumulation and reorganization of resource descriptors, belonging to 3 resources classes,
in a region of the Grid