Distributed computing to speed-up parametric studies
by M. Courtois, EDF R&D / AMA
The need of parametric studies has been increasing for some time : users want to run several jobs with simple variations of several parameters, for instance material properties, geometry, loads …
The whole number of computations can rapidly grow up to several thousands and, thus, its preparation and its submisssion could become a tedious work. Schematicaly, a parametric study is made up of a nominal study (a unique command file) and a set of varying parameters : Code_Aster is now able to automatically produce the set of individual studies, and to submit them taking into account the available nodes of computing resources. This availability of the compute nodes is directly connected to the speed-up : in this exemple below, 300 computations of 2 hours each are distributed on an hundred of processors. Then, the results are available in about 6 hours.
|
The management of parametric studies and their distributed submission is possible since the 1.8 version of Astk.
A new document [U2.08.07] « Distribution de calculs paramétriques » is describing the methodology for carrying out these studies with a minimum of intervention from the user.
This type of analysis is particularly efficient for the simulation of buildings under seismic loads. In the exemple below, we study the non-linear behaviour of a structure made of reinforced concrete. 300 computations are carried out with the combination of : 50 different seismic signals, variations of 3 material properties (Young’s modulus of concrete, damping, density).
|
|