- 10:10 AM
626e

Dynamic Cape-Open Simulation Appoach on Cluster Oriented Architecture

Laurent Pigeon1, Pascal Roux1, Bertrand Braunschweig1, and Thierry Gautier2. (1) Applied Mathematics and Computer Science, Institut Français du Pétrole, IFP-Lyon, BP3, 69 390 Vernaison, France, (2) Computer Science, INRIA / Laboratoire ID-IMAG, ENSIMAG - antenne de Montbonnot, ZIRST 51, avenue Jean Kuntzmann, 38330 Montbonnot Saint Martin, France

Based on the fact that today's simulations require clever strategies in order to obtain accuracy and performance, we have investigated a way to reduce the computational time of CAPE simulations. One of the methods that allow reducing the execution time of an application is to distribute the workload onto several computers also called cluster of PC. Based on INDISS software (RSI Dynamic Simulation Platform) and thanks to the CAPE-OPEN standard which provides a standard way to launch computations on Unit Operations, we have implemented a distributed layer on top of the INDISS runtime to execute CAPE-OPEN compliant process simulations. Using High Performance Computing techniques, our improved runtime can automatically launch and run a CAPE simulation onto several computers by scheduling (i.e., produce a placement of each Unit Operation component onto a logical partition), mapping each partition onto a physical processor, deploying the process configuration, managing the distributed execution and displaying results to the user.

There is two different ways to simulate a process: one may want to simulate a process to scale it (steady state simulation) or may want to study its behaviour during time (transient simulation). Our distributed layer manages as well steady-state as transient simulations.

In this paper, we present some techniques from parallel and distributed computing applied to CAPE-OPEN compliant process simulations. Based on experimental results on a deepwater application case, our approach is able to reduce at least by three the computational time of the transient or steady-state simulation while running on five PC computers.