|
Parallelized software : the current step to improve performances
|
|||
 |
||||
 |
 |
|||
Launch an application on multiple processors saves time in terms of calculations, each processor running a smaller part of the calculation. This division of tasks is now possible thanks to the architecture of computers and clusters.
However, to take maximum advantage of the benefits of this technology, the software has to be built on a such structure. This implies an additionnal and complex work but offering clearly superior performances.
In the field of numerical simulation, and in particular in the fluid mechanics, calculation performances are often discouraging.. Parallelized software and their use on powerful computing clusters are now as essential to modeling increasingly fine modelling and increasingly rapid calculations.
The ThosT and XimeX software, developed by SCC in collaboration with the CEMEF, are parallelized and fully operational in this mode of operation.
The Thost and XimeX software are parallelized. Thus, the tasks of automatic meshing and computations are distributed on the different available processors.
Through parallelization, in just one year, the number of processed items has considerably increased . As an example, the number of mesh elements on a twin screw modelling growths from 2 to 6 million tetrahedral elements, and on only 16 knots. Such an advantage makes it possible to refine the mesh on the most sensitive areas of the process for greater accuracy, and why not, to highlight previously invisible elements . Moreover, the computation times are quite reasonable to obtain results for the whole process treatment.
Indeed, the parallelization of the software allows to comutes the and reproduce the whole process, from the beginning until its end (which was unthinkable a few years ago).
This approach allows us to deal with specific and targeted elements of the process while taking into account the impact of the general behavior of the process on such critical elements .
This technology opens doors to new applications for numerical simulation ...
 |
|||
 |
|||