Dynamical computing power balancing for adaptive mesh refinement applications

This paper demonstrates a new computing paradigm for irregular applications in which the computational load varies dynamically and neither the nature nor location of this variation can be known a priori. The quality of the solution is improved through an iterative cell refinement process. The work load in different parts of the exploring domain may change based on the refinement criteria and the solution features of the system. In such instances, load imbalances between distributed processes becomes a serious impediment to parallel performance, and different schemes have to be deviced in order to balance the load. Usually, either data migration or data re-distribution techniques are deployed to solve the problem. In this paper, we describe a work which uses "dynamical computational power balancing". In this model, instead of balancing the load on distributed processes, the processes with heavier loads require the help from those with less loads. We illustrate the feasibility and practicality of this paradigm in an adaptive mesh application for solving non-linear dynamical systems via cell mapping method. The paradigm is illustrated using MPI for distributed memory programming and OpenMP for shared memory programming.