Reduced message count for interaction decomposition of N-body simulations

Abstract

Disclosed are a method of and system for creating a load balanced spatial partitioning of a structured, diffusing system of particles with pairwise interactions that is scalable to a very large number of nodes and has favorable communications characteristics including well defined bounds on the number of hops and the number of nodes to which a particle's position must be sent. To deal with structural imbalance, we can assign a weight corresponding to the computational cost for a particular pair interaction of particles or locally clustered groups of particles to simulation space at the midpoint of the distance between the particles or centers of the clusters. We then carry out a spatial partitioning of the simulation volume using a k-d tree or optimal recursive bisection (ORB) to divide the volume into sections that have approximately equal total weights. To deal with local degradation of the load balance caused by changes in the distribution of particles from that used to determine the original spatial decomposition one can assign the actual computation of the pair interaction between two particles to any node that has the positions of both particles.

Description

BACKGROUND OF THE INVENTION[0001]1. Field of the Invention[0002]This invention generally relates to methods and systems for solving N-body problems, such as classical molecular dynamics. More specifically, the invention relates to methods and systems for creating a load balanced spatial partitioning of a structured, diffusing system of particles.[0003]2. Background Art[0004]Over the last few years, unprecedented computational resources have been developed, such as the Blue Gene family of computers from the International Business Machines Corporation. These resources may be used to attack, among other issues, grand challenge life sciences problems such as advancing the understanding of biologically important processes, in particular, the mechanisms behind protein folding.[0005]In order to address this goal, attention has been directed to creating a classical molecular dynamics software package for long-time and large-scale molecular simulations. Classical molecular dynamics is predom...

AI Technical Summary

Benefits of technology

[0011]A further object of the invention is to enable the geometric n-body problem to be efficiently mapped to a machine with a fundamentally geometric processor space.

[0012]Another object of the invention is to create a load balanced spatial partitioning of a structured, diffusing system of particles with pairwise interactions that is scalable to a very large number of nodes and has favorable communications characteristics, including well defined bounds on the number of hops and the number of nodes to which a particle's position must be sent.

[0015]Because pair interactions are cut-off beyond a defined radius Rc, one can broadcast the position of a particle to the group of nodes containing a portion of the surface of a suitable specified convex shape that contains the sphere of radius Rb≧Rc / 2 centered about the position of original particle in simulation space, ensuring the existence of at least one node containing the positions of both particles required to compute any interaction. An example of such a convex shape is the spherical surface of radius Rb itself. The locus of points formed by the intersection of the two surfaces may be referred to as the “interaction loop”, and any node containing a portion of the “interaction loop” can compute the specified interaction. This provides an opportunity for eliminating imbalances caused by short-term local fluctuations in interaction workload. Because the load-balancing is carried out using interaction centers, there are many more geometrically distinct work objects that can be partitioned using an ORB strategy, and this decomposition allows productive use of more nodes than there are particles in the system.

Problems solved by technology

However for many scientifically relevant molecular systems, the computational burden due to the particle pair interactions remains large.

Images