Shared RAM Data Tables for Low-Memory Parallel Simulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing simulation methods for predicting pollutant emissions from vehicles with internal combustion engines face memory constraints due to large physicochemical tables, which exceed the capacity of computing machines, especially in parallel solvers using the MPI paradigm, leading to high RAM consumption and limiting the memory available for other simulations.
Innovation Solution
A computer system and method that utilize a single shared RAM segment for storing digital data tables, allowing all cores to access the data, reducing memory consumption and enabling efficient parallel calculations using an MPI message passing interface.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the entire table is loaded into memory for each parallel process, then the simulation can access all data, but the memory consumption increases significantly
Solution Approach 1:
The patent divides the large physicochemical table into multiple segments that are distributed across different memory spaces. Each parallel process only loads the specific segment it needs rather than the entire table, reducing memory consumption while maintaining data accessibility through coordinated access to segmented data structures.
Solution Approach 2:
The patent transitions from a flat memory loading approach to a hierarchical segmented structure, organizing data in multiple dimensions (process-level segments, memory-level segments, and file-level segments). This dimensional reorganization allows efficient data access without requiring full table loading in each process.
2Measurement precision
If large physicochemical tables are used to improve prediction accuracy, then the simulation precision increases, but the tables become incompatible with available memory capacity
Solution Approach 1:
The patent segments large physicochemical tables into manageable portions that can be stored in available memory. By dividing the comprehensive data tables into smaller segments distributed across multiple processes and memory locations, the system maintains access to complete high-precision data without requiring each individual process to hold the entire table in memory.
Solution Approach 2:
The patent introduces intermediary data structures and memory management layers that mediate between the large physicochemical tables and the limited process memory. These intermediaries enable efficient data retrieval and processing of high-precision tables without requiring full in-memory loading, allowing prediction accuracy to be maintained within memory constraints.
3Use of energy by moving object
If distributed memory architecture is used to reduce power consumption, then energy efficiency improves, but memory capacity per processor decreases
Solution Approach 1:
The patent merges the memory resources of multiple processes through shared segmented data structures. By combining access to distributed memory segments across processes, the system achieves effective access to large data tables while maintaining the low-power distributed memory architecture, as each individual processor still uses minimal local memory.
Solution Approach 2:
The patent creates universal segmented data structures that serve multiple processes simultaneously. These multi-functional data structures allow the same memory segments to be accessed by multiple parallel processes, maximizing the utility of limited per-processor memory while maintaining the energy-efficient distributed architecture.
Data Source
Figure 1~2
Figure 3~4
AI summary
The present invention relates to a system and method for predicting (simulating) a physical and/or chemical phenomenon on a plurality of processors (3), each processor having several processing cores (4). The system and method further comprise a random access memory (5) having a plurality of memory segments (6). The invention is based on the use of a numerical data table (2) stored in a single, so-called "shared" RAM segment (7) between all the cores (4).