Contact Pair Splitting for Parallel Finite Element Load Balance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed memory parallel processing systems face challenges in load balance during finite element simulation due to the presence of contact elements, which require more computations and can lead to uneven workload distribution among CPU cores, degrading simulation scalability and increasing solution times.
Innovation Solution
The system splits large contact pairs into smaller sub-pairs and distributes them across multiple CPU cores, using shared contact elements to ensure consistent solutions at splitting boundaries, and automatically determines the optimal split size to optimize load balance, employing different sharing methods based on contact detection algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all contact elements in each contact pair are restricted to a single core, then a convergent solution is obtained, but the load balance deteriorates and solution times increase
Solution Approach 1:
The contact pair is segmented into multiple contact sub-pairs by dividing the contact surface into discrete elements. Each contact sub-pair can be assigned to different CPU cores, allowing parallel processing while maintaining convergence through shared contact elements at the boundaries between sub-pairs.
Solution Approach 2:
Contact elements are nested across multiple hierarchical levels: individual contact elements form contact sub-pairs, which are grouped into contact pairs, and multiple contact pairs are distributed across CPU cores. This nested structure enables parallel processing at the core level while maintaining sequential consistency at the contact pair level.
2Productivity
If contact pairs are distributed across multiple CPU cores, then parallel processing efficiency improves, but load balance deteriorates due to uneven computational workload
Solution Approach 1:
Different regions of the contact surface are treated differently based on their computational requirements. Contact elements near boundaries share data with adjacent sub-pairs, while interior elements use standard computation. This local differentiation optimizes the computational workload distribution across cores.
Solution Approach 2:
The computational parameters are dynamically adjusted based on the distribution of contact elements. The system automatically determines the optimal number of splits for the largest contact pair and proportionally adjusts other contact pairs to achieve balanced workload distribution across available CPU cores.
3Adaptability or versatility
If the number of contact pairs increases, then the model complexity increases, but scalability deteriorates due to restricted contact element assignment
Solution Approach 1:
The contact pair splitting strategy is dynamic and adapts to the specific model characteristics. The system automatically determines the optimal split configuration based on the number of contact pairs and available computational resources, enabling the model to scale efficiently as complexity increases.
Data Source
AI summary
Methods for modeling contact pairs in a model of a physical object include generating a contact pair including a contact surface and a target surface, where the contact pair further includes contact elements of the contact surface and the target surface, splitting the contact pair into contact sub-pairs along splitting boundaries, augmenting each contact sub-pair with contact elements from adjacent contact sub-pairs at the splitting boundaries, distributing the augmented contact sub-pairs to a plurality of parallel processors for finite element solutions of the contact sub-pairs, receiving the finite element solutions of the contact sub-pairs from the plurality of parallel processors, and combining the finite element solutions of the contact sub-pairs into finite element solutions of the contact pair.