Information Technology Reference
In-Depth Information
18
16
14
12
10
8
6
4
2
0
1000
2000
3000
4000
5000
10000
Numberofagents
Noconstraints
Withdummyconstraint
Figure 9.14:
Results of benchmark suite 2.
leads to a significantly steeper growth of runtime. This effect is more
evident if more agents are present in the model.
In theory, constraints lead to a quadratic growth of runtime (see
Section 8.1.2). The benchmarks with up to 10000 agents indicate
that (by regression analysis) the runtime depends on the number of
agents in the following way:
t
r
=0
.
08
n
2
A
+0
.
91
n
A
+0
.
034. Of course,
these coecients are highly dependent on the chosen model and the
implementation of the specific simulation engine. Furthermore, the
coecients are highly sensitive to, at least, two factors:
Time for evaluating a constraint
The most influential impact factor is the time needed for a single
evaluation of a constraint. Within the presented benchmark suites,
the dummy constraint contains no actual logic, but only a number
of loops and string operations in order to consume time. The more
complex a constraint is, the more computation and data is necessary
to evaluate a constraint. Assuming
O
(
n
2
) constraint evaluations,
A
