In contrast to its counterpart, a top-level parallelization strategy
(Figure 6.2) is splitting up a simulation
problem at the highest possible level. Let us consider a design of
experiments involving several independent executions of a
simulator. Instead of executing a bottom-level parallelized simulator
sequentially, top-level parallelization executes several
(sequential) simulators in parallel.
Since individual executions of a simulator are independent from each other, a communication overhead does not exist at all. Moreover, top-level parallelization does not require any modifications to an existing simulation tool. As long as there exists software which is able to deal with remote execution and the synchronization of the simulations, top-level parallelization can easily be applied to arbitrary simulation tools. It should be stressed that this kind of parallelization is especially efficient when applied to ``design of experiments'' applications or optimization problems. Since these tasks tend to require large numbers of simulator invocations resulting in huge amounts of computation time, a good scalability is highly desirable.