Multiprocessor Scheduling of Simulation Code From Modelica Models
Peter Aronsson, Peter Fritzson · 2002
Modern object oriented modeling techniques, such as the Modelica modeling language, are increasing the capability to model and simulate systems of large size and complexity. Simulation of such large and complex systems is computationally very expensive. The use of parallel computers for simulation of Modelica models is one approach of handling simulation of such large and complex systems within reasonable time limits. This paper presents an automatic parallelization tool that translates the sequential simulation code generated from a Modelica compiler, Dymola, into a parallel version that can be executed on a parallel computer. The paper also presents several scheduling and clustering techniques used by the tool to partition the simulation code onto several processors. One of these techniques, called FTDT-Full Task Duplication Technique, gives a measured speedup of 2.5 on an 8 pro- cessor PC-cluster. However, future work includes de- veloping better scheduling and clustering algorithms to further improve the results.