| | |
Summary: On the Performance of a Distributed Object
Oriented Adaptive Mesh Refinement Code
William K. Allard \Lambda John A. Trangenstein y
November 26, 1997
1 Introduction
In this paper we describe how we modified the adaptive mesh refinement (AMR)
code described in [4] so that it would run efficiently on parallel machines and
we give some statistics about its performance on the Cray T3E. Since the first
target architecture for the modified code was the Cray T3E, which is a dis
tributed memory machine, we call this modification Distributed Adaptive Mesh
Refinement (DAMR).
AMR codes are well known to be large and complex. What's more, given
the scientific context in which they are used, one is always incorporating addi
tional features to expand their utility. One effective way to deal with program
complexity and maintenance problems is to use object oriented programming
to manage the higher level data structures in the code. When running AMR
code on a parallel machine, program complexity increases even more because
communication between hosts needs to occur in a manner which is determined
dynamically. It is not at all clear how to keep the processors busy while at the
same time synchronizing them so as to allow the communication to take place
|
