Towards autonomic application-sensitive partitioning for SAMR applications

Abstract

Distributed structured adaptive mesh refinement (SAMR) techniques offer the potential for accurate and cost-effective solutions of physically realistic models of complex physical phenomena, and can provide new insights into applications such as interacting black holes and neutron stars, formations of galaxies, subsurface flows in oil reservoirs and aquifers, and dynamic response of materials to detonation. However, these applications are heterogeneous and dynamic in nature, and when combined with a similarly heterogeneous and dynamic execution environment such as the computational Grid, result in significant runtime management challenges. In this paper, we investigate autonomic runtime management approaches to enable the efficient and scalable execution of SAMR applications in Grid environments. This paper presents the design, implementation and evaluation of ARMaDA, a self-adapting and optimizing partitioning framework for SAMR applications. It then selects, configures, and invokes appropriate partitioning and scheduling mechanisms to match the current state of the application and optimize its computational and communication performance. The advantages of the autonomic partitioning capabilities provided by ARMaDA are experimentally demonstrated using a selection of SAMR kernels and different computational environments.

Files

Additional details