Aspects of coordination methods

In this section, we compare the three coordination methods defined in Section 5.3.1. As a measure for comparison, we look at the following five aspects:

• Use of preceding information

• Coordination complexity

• Simulated system designs

• Flexibility

• Throughput time

For every aspect, after defining it, we discuss the performance of the three coordination methods on this aspect and compare the methods. Finally, in Section 5.3.3, we summarize the results and give recommendations on the choice of a coordination method.

Use of preceding information: This aspect refers to the use of simulation and modeling results

from preceding black boxes in the decomposition graph. These results can help in the determination of combinations of design parameters that are expected to yield the most valuable information. Note that in case of a connection among design parameters and response parameters, the use of preceding information is a necessity to obtain simulated system designs. Clearly, both sequential coordination methods use preceding information by means of the response input parameters. In parallel simulation, no preceding information is used.

Coordination complexity: Coordination complexity refers to the amount of communication and time

that is needed to implement a coordination method. It includes costs that are incurred by, e.g., the need for an automated communication system. Often part-level problems are solved by several design teams. When the parallel simulation coordination-strategy is used, every design team operates independently and there is no need for a complex organizational structure; see, e.g., Krishman [1996], where managing the simultaneous execution of two coupled development phases plays a central role. In sequential simulation, communication is needed after each evaluation of a black box. Therefore, this coordination method results in a complex coordination process that needs

complete design has been run. Hence, the coordination process is relatively simple compared to sequential simulation.

Simulated system designs: A system design is a particular setting of all the design parameters in

the system. When we simulate such a setting, we obtain a simulated system design or system-level simulation. However, the simulations are carried out by simulation tools at the part-level; each tool has a subset of the system design parameters as input. Further, linking design parameters and design parameters that are black box outputs create overlap in the sets of design parameters and couple the black boxes. In order to obtain a simulated system design, we must use the same setting for the linking design parameters at every black box and use the simulation results of response input parameters as design parameter settings in successive simulation tools. Clearly, it is not possible to obtain simulated system designs in parallel simulation, because all linking design parameters are treated independently. Simulated system designs can be obtained in sequential simulation and sequential modeling, since these coordination methods use the simulation results from preceding simulation tools as inputs. Besides, it is also necessary to use the same settings for the linking design parameters in all simulation tools. This can be accomplished by the construction of coupled designs (see Husslage et al. [2005]). Obtaining simulated system designs may require some effort, but it is a great help in the system design process. After step 2, promising designs may have been found even without using the meta-models.

Flexibility: Flexibility of a coordination method means that it does not take much effort to validate

or adjust the constructed meta-models, when a small change is made to one or more simulation tools. Meta-models are based on results found by the black boxes. Hence, should a simulation tool be adjusted, (e.g., due to changes in the underlying component) then the constructed meta-model will probably no longer be valid. In parallel simulation, this can be fixed by simulating an extra set of designs for the corresponding component, since the meta-models for the various black boxes are constructed independently. However, in the two sequential coordination methods, one must be more careful, since coupling among simulation tools is preserved in the construction of meta- models. Therefore, invalidity of one meta-model can affect the validity of the meta-models of all its successors. Since small changes may require much effort in the validation and, possibly, adjustment of many meta-models, the sequential coordination methods are not flexible with respect to changes in the simulation tools, whereas parallel simulation is flexible.

Throughput time: The throughput time of a coordination method is defined as the total required

chain. From a time-to-market perspective, it is desirable to have short product development times, so the throughput time should preferably be small. Since the construction time of meta-models is assumed to be negligible relative to the simulation run time, we ignore it in this analysis. Parallel simulation always leads to the shortest throughput time; sequential simulation takes more time; sequential modeling takes the most time. One can simply compute the exact throughput time for each coordination method in a specific problem instance (see the formulas in Husslage et al. [2003]). This information, along with the four other aspects, can be used to make a decision about the coordination method to use.

Suppose for example that the simulation times for the black boxes in the example of Figure 21 are given by Table 13, and we need to run 40 simulations for each black box. Then, the throughput time for parallel simulations is 2,400 min., for sequential modeling 2,438 min., and for sequential simulations 3,520 min.

Black box Simulation time

1 60 min. 2 30 min. 3 25 min. 4 35 min. 5 60 min. 6 3 min.

Table 13. The simulation times for the black boxes of the example from Figure 21.