Information Technology Reference
In-Depth Information
would be no modification of the update cost of the environment. However, when the
matching is executed by agents, the number of messages in the update process would
be increased by
freq
a
(
n
a
−
1)
(the cost of broadcasting the description updates), thus
improving the comparative environment performance.
In this section the assessment has been done with the hypothesis that the agents have
the same information to evaluate a filter. In the next section, we depict how these results
relate to real experimentation where this hypothesis is not valid.
5
Empirical Assessment
In section 4, we have shown that using the environment to compute the contexts is less
costly than computing them locally in the agents when the agents update frequency is
not too high. In other words, the entities dynamics is the parameter that determines
which solution is the best for a multi-agent system. This section consists in evaluating
the cost of the context computing on a real example that is a simulation of the victim
evacuation in the crisis situation.
A prototype of our ABS framework has been implemented as a plugin for the multi-
agent platform Madkit [7]. This plugin is composed of an environment component with
an API that enables the agents to add/retract/modify their descriptions and filters. We
have chosen to implement the matching process within a Rules-Based System (RBS).
The instantiation of the model into a RBS is straightforward: the descriptions are the
facts of the rule engine, and the filters are its rules. Rule firing is based on the efficient
R
ETE
algorithm [6]. It is a network-based algorithm designed to speed the matching
of patterns with data. RETE uses a static discrimination network, generated by the lan-
guage compiler, that represents data dependencies between rule conditions.
We compare our model which encompasses contextual activation by the environment
with a model which encompasses a classical activation process to evaluate the cost
of supporting the simulation process through the environment. We have tested four
simulation scenarios, using ten filters: seven filters for contextual activation (
f
1
to
f
7
),
two filters for communication (
f
recept
and
f
accept
). The filter
f
classicalActivation
is
used in the scenarios
S
1
and
S
3
in order to simulate a classical activation process. It
activates each agent once by simulation cycle without taking into account the context.
The other activation filter allows to activate an agent
medical porter
in a specific context
to perform an action.
The scenarios are defined below.
S
1
and
S
2
are scenarios without agent communica-
tion, and
S
3
and
S
4
are scenarios with agent communication:
-
-
S
1
-
{f
classical activation
}
+ Local Agent Context Analysis (LACA),
-
-
S
2
-
{
f
1
,f
2
,f
3
,f
4
,f
5
}
,
-
-
S
3
-
{
f
classical activation
}
+ LACA + Broadcast,
-
-
S
4
-
{
f
1
,f
2
,f
3
,f
4
,f
5
,f
6
,f
7
}
+
{
f
recept
,f
accept
}
S
1
illustrates a classical scenario with an activation phase and a local agent context
analysis;
S
2
is a scenario with a contextual activation inside the environment. We de-
scribe the filters of
S
2
. The filter
f
1
allows to activate the action
move randomly
that is
triggered when a medical porter has no victim close to it (context
victim seeking
). The
