High-Level Modeling and Exploration of a Powerline Communication Network Based on System-on-Chip - Multi-objective Design Space Exploration of Multiprocessor SoC Architectures - page 160

Table 7.6 Semi-automated
procedure steps

STEP Description

Step 0 Build a configurable platform using ASCII files

(DS2 proprietary lan-guage) to configure the
platform. Prepare the platform to overwrite some
of these configurations by MULTICUBE own
tools

Step 1' Automated parameter update procedure and metric

extraction

Step 2' Final analysis of the solutions in order to select the

optimized combination of parameters

￿ Semi-automated procedure (or “traditional procedure”) . In order to optimize

different chipsets, DS2 has developed an in-house semi-automated procedure.
This procedure is possible thanks to the high flexibility and configurability of the
STORM platform. Using this platform, the optimization procedure is performed
by following the steps shown in Table 7.6 .

￿ Automated procedure (MULTICUBE flow) . MULTICUBE procedure has been

applied to DS2 use case in order to optimize the above mentioned parameters.
As a consequence, the number of steps have been reduced significantly. The
resulting flow is presented in Table 7.7 . The procedure has been run on a dual-
core desktop PC with Ubuntu 8.04 OS for 36 h under the supervision of a team of
two engineers. Thanks to MULTICUBE automation, up to 534 experiments have
been done without human intervention. In order to achieve this result we have
setup the experiments using the elements of the modeFRONTIER workflow (Fig.
7.8 ). In this workflow, we have selected the points to analyze in two ways: one by
choosing an initial Design of Experiments (DoE) composed of a random sampling
and another by using MOGA-II algorithm (Multi Object Generic Algorithm) as

Table 7.7 Automated procedures steps
STEP

Description

Step 0 Build a configurable platform using ASCII files (DS2 proprietary lan-guage) to

configure the platform

Step 1 Select a subset of 2-3 parameters and 2-3 metrics

Step 2 For the non-selected metrics, fix the rules of what are the accepted boundaries

Step 3 Write automation scripts to generate all selected combinations and to filter the

requested metrics from the output files and facilitate manual in-spection

Step 4 Run simulations for all possible selected combinations with the automated script and

automatically extract metrics from the simulation results

Step 5 Manual inspection of metrics. Selection of a set of the parameter com-bination that

seems to be the better adapted. Selection is done through designer's experience

Step 6 Generate new platforms with the selected combination of parameters and corner cases

for the non-selected parameters

Step 7 Run simulations with the new set of platforms

Step 8 Verify that the ALL the metrics meet the rules specified in step 2. If not, Investigate

causes. If the cause can be corrected by changing one of the selected parameters
select a new combination in step 5. If the cause cannot be corrected by changing one
of the selected parameters Change the selected combination of parameters (include
new ones) and come back to step 1