Parallel Processing System Local Memory Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

In parallel processing systems, shared memory type systems face performance issues due to frequent access competition among computers, while distributed memory type systems require complex synchronization processes, which hinder efficient processing performance.

Innovation Solution

The system employs a local memory group with a first and second area for each computer, switching access between these areas periodically, allowing processes to execute independently without the need for synchronization, thus eliminating access competition and simplifying control functions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If a shared memory type parallel processing system is used, then memory access is simplified, but access competition occurs frequently reducing process performance

Engineering Contradiction:
Improvememory access simplicityVSAvoidprocess performance
Core Design Contradiction:
Ease of operationVSProductivity

Solution Approach 1:

The memory system is segmented into a shared memory section and separate local memory sections (first and second local memory areas) for each computer. This segmentation allows each computer to access its own local memory independently while still enabling shared memory access when needed, thereby eliminating access competition and improving process performance while maintaining ease of operation.

Inventive Principle:
Principle #1Segmentation

2Productivity

If a distributed memory type parallel processing system is used, then access competition is eliminated, but complex synchronization control functions are required

Engineering Contradiction:
Improveaccess efficiencyVSAvoidsynchronization control complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The synchronization control function is extracted from each computer and consolidated into a dedicated synchronization control section. This externalizes the complex synchronization logic, allowing individual computers to execute processes independently without embedded synchronization complexity, while the centralized synchronization control section manages data consistency across all computers.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The synchronization control section acts as an intermediary between the distributed computers and the shared memory system. It mediates data synchronization by copying data between local memory areas and shared memory as needed, thereby eliminating the need for complex synchronization control functions in each computer while maintaining data consistency.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Reliability

If synchronization process is implemented in distributed memory system, then data consistency is achieved, but process execution must be stopped reducing performance

Engineering Contradiction:
Improvedata consistencyVSAvoidprocess execution continuity
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

Data is preliminarily prepared in separate local memory areas before being transferred to shared memory. The first and second local memory areas serve as pre-processing buffers where data can be prepared independently and concurrently. This preliminary action in local memory areas enables subsequent synchronization to proceed without stopping process execution, as data is already ready for transfer.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system maintains continuity of useful action by allowing process execution to proceed independently in each computer while the synchronization control section continuously manages data synchronization in the background. The use of dual local memory areas enables overlapping of computation and synchronization operations, ensuring that data consistency is maintained without interrupting the continuous execution of processes.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP2693343B1Parallel processing system and parallel processing system operation method
Publication Date: 2019.02.13 MITSUBISHI HEAVY IND LTD
  • EP2693343B1 patent drawingFigure 1A
  • EP2693343B1 patent drawingFigure 1B
  • EP2693343B1 patent drawingFigure 2

AI summary

A parallel processing system includes a plurality of computers accessibly connected through a network, and distributedly executing a plurality of processes. Each of the plurality of computers is composed of an operation processing unit configured to execute an allocated process, a local memory group having a first area and a second area and an I/O control circuit. The operation processing unit executes the allocated process by using the first area as an access destination in a first period and, and executes the allocated process by using the second area as the access destination in a second period subsequent to the first period. The I/O control circuit is composed of an updating section configured to update data stored in the local memory group to the latest data by carrying out communication among the computers. The updating section updates the data stored in the first area in the second period.