Parallel methodologies for large-scale simulation

摘要

System simulation involves the construction of a computationalnanalogue of a physical system-a process which has traditionally beennassociated with intemperate demands on processing time. This is morentrue today than ever before. Complex large-scale plant simulators andnprocess mimics are almost commonplace in some high-technologynindustries. In such settings, simulation may be used in both the designnand validation of novel control strategies, as well as forming part ofnan operator training programme, or in cause-consequence diagnostics. Thenproblem arises that as physical systems evolve in complexity, thendimensionality of the underlying dynamical equations can risendramatically, creating an immense computational burden at simulationntime. Often this may result in the violation of any real-timenconstraints, and the effects may become so acute as to render computerntimes that are excessive to the point of impracticality. Consequently,nthe authors are concerned with the development of parallel methodologiesnfor such problems. Attention is focused on a necessarily restricted setnof large-scale systems (LSS): those that are adequately described bynhigh-dimensional sets of ordinary differential equations (ODE's) innstate-space form. It should be noted however, that many of the generalnconcepts relating to parallelism exploitation, extend to a much widernclass of problems