@InProceedings{Supelec908,
author = {Virginie Galtier and Stephane Vialle and Cherifa Dad and Jean-Philippe Tavella and Jean-Philippe Lam-Yee-Mui and Gilles Plessis},
title = {{FMI-Based Distributed Multi-Simulation with DACCOSIM}},
year = {2015},
booktitle = {{Symposium on Theory of Modeling and Simulation, TMS'15}},
pages = {804-811},
month = {April},
address = {Alexandria, VA, USA},
abstract = {Our research project aims at enabling multi-simulation based
on the FMI 2.0 standard and the cooperation of multiple
FMUs (FMI simulation units). In order to support large scale
multi-simulations, our solution (DACCOSIM) runs on multi-
core and distributed architectures. To support variable step
size, the necessary error control and rollbacks are achieved
through a hierarchical and distributed control architecture. At
each step, simulation data communications also occur, but di-
rectly between FMU pairs in a fully decentralized fashion.
Moreover, DACCOSIM implements an algorithm to perform
the complex initialization of the various components of the
multi-simulation. DACCOSIM comes as a graphical frame-
work to easily design a multi-simulation and to automatically
generate associated code, and as a multithreaded and dis-
tributed library to execute it. We evaluated DACCOSIM on
an industrial use case provided by EDF (leading French util-
ity company), run on multi-core PCs and PC clusters. Pre-
liminary performance measurements on a 4-physical-core PC
exhibit a speedup compared to monothreaded Dymola execu-
tion using the same FMUs. On multi-core PC clusters we face
overhead communication times due to frequent small com-
munications but this distribution allows to process large co-
simulations.}
}