Introduction
The reason for the development of the Parallelizer is the ability to simultaneously launch various instances of Atelier B in order to accelerate the processing of proof that can very rapidly require exponential calculation time as a project becomes increasingly complex. This study therefore looks in depth at the time-saving the Parallelizer provides through the analysis of the time required for proof at a “relaunch” level (-2) for a project of industrial scope.
Project Used
The project used to conduct this comparison is relatively large and in fact includes 285 components. The processing time for this type of “relaunch” can be expressed in minutes, even with modern, effective computerized means.
Results
The first test is based on the use of the Parallelizer on a single machine with a single, one-time launch of Atelier B, therefore a “classical use” of the software, and somewhat acts as the control arm of the study.
The second test is a repeat of the prior one, however it authorizes a second launch on the same machine as the one used for the first launch. Its objective is to test the benefits of multi-processor machines.
The third test corresponds to the simultaneous launch of Atelier B three times on three different machines. This third test therefore looks at the benefits of distributing tasks over a network.
Finally, the last test involves two simultaneous launches on two distinct machines, which results in four simultaneous Atelier B launches.
These choices of test equipment use are based on the need to compare, first, the benefit of parallelizing the execution of tasks and, second, the effectiveness of machines with multiple processors/cores as compared to a simultaneous use of multiple machines.
The relaunch results (level -2 proof) further to a complete omission of proof, according to the various cases referred to above, are as follows:
| Test | Total number of launches | Number of machines | Time required (in minutes) |
|---|---|---|---|
| 1 | 1 | 1 | 29 minutes |
| 2 | 2 | 1 | 13 minutes |
| 3 | 3 | 3 | 11 minutes |
| 4 | 4 | 2 | 08 minutes |
Interpretation of the results
We note that the higher the number of simultaneous launches, the more rapidly the test is executed, which is logical.
However, we also note that the gain in execution onmulti-core machines is to be favoured as compared to a distribution over multiple single-processor machines, as is shown by the difference in results between Test No. 3 and Test No. 4.
In all cases, the use of the Parallelizer seems quite beneficial in the context of the development of relatively large industrial projects.
English
French 
