Postdoctoral Fellow Dr. M. Sipior is working on the integration of a direct N-body code with a Barnes-Hut treecode. The main goal of this effort is to acquire a computational environment which has, in combination with the hitherto described computer, sufficient dynamic range that clusters with a large number of stars can be simulated with the required accuracy. The end result allows us to make use of the unique advantages of each numerical approach--the high precision of the direct N-body integrator for studying the dynamics of a large stellar cluster, and the computational speed of a tree code, used to simulate the galaxy in which the cluster is embedded. The first stage of this effort is nearing completion; we are currently commissioning the code and preparing a publication with the numerical implementation.
Preliminary tests reveal that the break-even point in performance between a treecode running on an 8-node Beowulf and a direct N-body code running on a single node of MoDeStA is reached when running with a ratio of about 10-to-1 particles (ten times more praticles for the treecode than for the direct code).
PhD student A. Gualandris is working on parallelizing N-body kernels.
Since MoDeStA is a parallel platform (see above), we require an
efficient parallel N-body code to mine the machines' power. The first
results of our parallelization effort can be found at
http://carol.science.uva.nl/
spz/act/modesta/Software/index.html
At this site we also make the source code publicly available.
The first steps in making our parallel N-body software GRID operable have been taken (work of Dr. A. Tirado [UvA] and A. Gualandris [UvA]), and we have results on our first tests for a direct N-body calculation on a distributed computer. A publication about this work is in preparation.