ARMaDA: An Adaptive Application-sensitive Partitioning Framework for SAMR Applications

S. Chandra and M. Parashar (USA)


Adaptive application-sensitive partitioning, Dynamic SAMR applications.


Distributed implementations of dynamic adaptive mesh refinement techniques offer the potential for accurate solutions of physically realistic models of complex physical phenomena. However, configuring and managing the execution of these applications presents significant challenges in resource allocation, data-distribution and load balancing, communication and coordination, and runtime management. This paper presents the design and evaluation of the ARMaDA framework for adaptive application sensitive partitioning of dynamic structured adaptive mesh refinement applications. The ARMaDA framework has three components: application state characterization component, partitioner selection component, and adaptive meta-partitioner component that dynamically configures partitioning strategies at runtime based on current application state. Experimental results show that adaptive application-sensitive partitioning using the ARMaDA framework can improve application performance as com pared to non-adaptive partitioning.

Important Links:

Go Back