Particle Simulation Load Balancing via Region Assignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional particle simulation methods face inefficiencies in parallel processing due to unequal calculation loads among regions, particularly when the shape of the particle distribution is complex, leading to increased calculation time and difficulty in equalizing particle numbers across regions.
Innovation Solution
The particle simulation apparatus assigns regions to processes in descending order of particle number, with unassigned regions sequentially assigned based on the particle count of already assigned regions, ensuring that regions with higher loads are initially assigned to processes and balancing loads across processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If regions are divided such that the number of particles in each region is equalized, then calculation loads are equalized, but this becomes difficult when the shape of particle distribution is complex
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing the number of particles in each region before the actual simulation begins. This preprocessing step creates a particle number database that guides subsequent region assignment, allowing the system to handle complex particle distributions without recalculating during simulation runtime.
Solution Approach 2:
The patent changes the assignment parameter from simple sequential or random assignment to assignment based on particle number magnitude. By sorting regions in descending order of particle count and assigning them strategically to processes, the system achieves better load balancing even when particle distribution shapes are complex and irregular.
2Loss of time
If conventional region assignment methods are used, then implementation is simple, but calculation loads become unequal leading to increased calculation time
Solution Approach 1:
The patent implements feedback by continuously monitoring the particle count in each region and using this information to dynamically adjust region assignment to processes. The system calculates particle numbers, compares them across regions, and assigns regions to processes based on this feedback, ensuring that processes with fewer particles receive additional regions to balance the load.
Solution Approach 2:
The patent applies dynamics by making the region assignment process adaptive rather than static. Instead of fixed assignment, the system dynamically adjusts which regions are assigned to which processes based on real-time particle count information, allowing the parallel processing system to optimize its performance according to the actual particle distribution.
3Ease of operation
If regions are assigned sequentially without consideration of particle number, then assignment is straightforward, but load balancing is poor
Solution Approach 1:
The patent applies segmentation by dividing the region assignment process into distinct phases: first sorting regions by particle number in descending order, then sequentially assigning them to processes while monitoring load balance. This segmented approach maintains operational simplicity through automated sorting and assignment while achieving superior load balancing compared to unsorted sequential assignment.
Data Source
AI summary
When simulating behavior of particles in a space having regions and subject to analysis, by processes of a predetermined number fewer than the number of regions, a particle simulation apparatus assigns the regions of the predetermined number, which are selected in descending order of the number of particles included in each of the regions, to differing processes among the processes of the predetermined number. The particle simulation apparatus sequentially assigns unassigned regions in descending order of the number of particles included in each of the unassigned regions, to the process for which the number of particles included in the regions already assigned to each of the processes of the predetermined number is the smallest.