Dynamic Region Boundary Adjustment for Particle Simulation Load Balance

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional particle simulation methods using parallel computers face inefficiencies in resource allocation when dealing with multiple types of particles having different calculation costs and distribution patterns, leading to suboptimal load balance and decreased parallelization efficiency.

Innovation Solution

A particle simulation program that assigns partial regions to nodes based on the calculation cost and number of processing-target particles of each type, dynamically adjusting the region boundaries to equalize calculation amounts across nodes, thereby optimizing resource allocation and improving parallelization efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If particles are distributed uniformly across nodes without considering calculation cost, then the region division is simple, but the load balance deteriorates

Engineering Contradiction:
Improveregion division complexityVSAvoidparallelization efficiency
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent applies local quality by assigning particles with different calculation costs to different nodes based on their specific computational requirements. Instead of uniform distribution, each node receives a tailored allocation of particle types and quantities matched to its computational capabilities, ensuring that nodes handling more complex calculations receive appropriate resource allocation while maintaining overall system efficiency.

Inventive Principle:
Principle #3Local quality

2Device complexity

If region boundaries are fixed without dynamic adjustment, then the allocation is simple, but the load balance deteriorates when particle distribution changes

Engineering Contradiction:
Improveboundary adjustment complexityVSAvoidload balance
Core Design Contradiction:
Device complexityVSProductivity

Solution Approach 1:

The patent implements dynamics by making region boundaries adjustable rather than fixed. The system dynamically modifies boundary positions between partial regions in response to changing particle distributions and calculation costs. This allows the computational domain to adapt to evolving simulation conditions, maintaining optimal load balance across nodes even as particle concentrations and computational requirements change during the simulation process.

Inventive Principle:
Principle #15Dynamics

3Ease of operation

If all nodes execute the same number of particles regardless of type, then the distribution is uniform, but the calculation load becomes unbalanced

Engineering Contradiction:
Improveparticle distribution uniformityVSAvoidcalculation load balance
Core Design Contradiction:
Ease of operationVSProductivity

Solution Approach 1:

The patent applies parameter changes by varying the allocation parameters of particles assigned to each node based on their calculation costs and the specific computational requirements of different particle types. The system adjusts the number and types of particles distributed to each node, transforming the uniform distribution approach into a parameterized allocation strategy that optimizes computational load balance while maintaining operational simplicity.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS10503561B2Particle simulation apparatus and computer resource allocating method
Publication Date: 2019.12.10 FSAS TECH INC
  • US10503561B2 patent drawing
  • US10503561B2 patent drawing
  • US10503561B2 patent drawing

AI summary

A parallel computer that includes a plurality of nodes assigns to each of the nodes a partial region that is a division of a region in which a plurality of types of particles are distributed, and executes a plurality of programs for a particle simulation by each of the nodes. Then, according to a type of a processing-target particle of each of the plurality of programs and an execution time of each of the plurality of programs, the parallel computer determines a computation cost for each of a plurality of processing-target particles of each of the plurality of types. Subsequently, the parallel computer changes a position of a region boundary of the partial region according to the computation cost and the number of the processing-target particles of each of the plurality of types.