171ADAPTIVE MESH REFINEMENT A PARALLEL ADAPTIVE MESH REFINEMENT ALGORITHM FOR SOLVING NONLINEAR DYNAMICAL SYSTEMS

An unstructured adaptive mesh refinement (AMR) method is used in conjunction with the cell-to-cell mapping method for solving nonlinear dynamical systems. The global anal-ysis is initiated with a coarse mesh without any a priori knowledge of the cell state space of the system. The process of investigation of nonlinear systems is improved through an iterative cell refinement and cell lumping proc-ess. In this paper, a parallel algorithm for the AMR proce-dure is developed using a shared memory programming paradigm. In addition, the concept of dynamic computing for better resource management is introduced. The approach compensates for changes in the computing workload by dynamically allocating/deallocating shared memory threads for more efficient computations. Results on the SGI Origin series are presented to exhibit the scalability of the algo-rithm and the feasibility of dynamic computing for resource management