Memory Analysis Device Optimizing DCPMM and DRAM Configurations

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems lack an efficient method to determine optimal hardware configurations for applications using DCPMM, balancing performance and cost, as DCPMM is not suitable for all applications due to varying memory access patterns and requirements.

Innovation Solution

An analysis device and computer system that evaluates memory performance characteristics using benchmark applications and determines recommended memory configurations by analyzing memory reservation and access amounts, recommending configurations that optimize cost and performance based on specific application patterns.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If DCPMM is used for memory capacity expansion, then cost is reduced, but performance deteriorates compared to DRAM

Engineering Contradiction:
Improvememory capacityVSAvoidmemory access speed
Core Design Contradiction:
Quantity of substanceVSSpeed

Solution Approach 1:

The system changes the parameter of memory type from DRAM to DCPMM based on the determined memory access characteristics. When applications exhibit sequential access patterns or large memory reservation with low access frequency, the system configures DCPMM instead of DRAM, accepting slower speed in exchange for lower cost and adequate performance for these specific access patterns.

Inventive Principle:
Principle #35Parameter changes

2Ease of operation

If DCPMM is used in memory mode, then no application modification is needed, but simultaneous installation of DRAM is essential increasing complexity

Engineering Contradiction:
Improveapplication compatibilityVSAvoidmemory configuration complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The system determines the appropriate memory mode (memory mode or App direct mode) and configuration parameters based on analyzed application characteristics. For applications with large memory reservation and sequential access patterns, the system configures DCPMM in memory mode with DRAM as cache, while for other patterns it may recommend App direct mode or different DRAM configurations, thereby optimizing the balance between ease of operation and device complexity.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If hardware configuration is optimized manually by system designers, then configuration accuracy improves, but time consumption increases

Engineering Contradiction:
Improveconfiguration accuracyVSAvoidconfiguration time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs automatic analysis of application memory access characteristics by executing benchmark applications and measuring memory reservation and access amounts. Based on this automated measurement, the system independently determines the optimal memory configuration without requiring manual intervention from system designers, thereby maintaining high configuration accuracy while significantly reducing the time required for hardware configuration optimization.

Inventive Principle:
Principle #25Self-service

4Speed

If extensive hardware changes are made to optimize performance, then performance improves, but cost and complexity increase

Engineering Contradiction:
Improvememory access performanceVSAvoidhardware configuration complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The system analyzes application memory access patterns and determines optimal memory configurations by changing parameters such as memory type (DRAM vs. DCPMM), memory mode (memory mode or App direct mode), and DRAM/DCPMM combination ratios. This parameter-based optimization achieves performance improvement without requiring extensive hardware changes, thereby reducing both cost and configuration complexity while maintaining or enhancing memory access performance.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS11301133B2Analysis device, non-transitory computer-readable storage medium for analysis program, and computer system
Publication Date: 2022.04.12 FUJITSU LTD
  • US11301133B2 patent drawing
  • US11301133B2 patent drawing
  • US11301133B2 patent drawing

AI summary

An analysis device configured to be connected to an information processing apparatus configured to mount a first memory and a low-speed second memory, the low-speed second memory being cheaper and having lower performance than the first memory and being used for memory capacity expansion, the analysis device being configured to perform program instructions including: causing the information processing apparatus to execute a plurality of types of performance evaluation application programs and acquire memory performance characteristic information regarding each performance evaluation application program from the information processing apparatus; determining a recommended memory configuration according to the performance evaluation application program corresponding to the application to be evaluated program among the plurality of types of performance evaluation application programs by using a collection result of the memory performance characteristic information; and outputting recommended memory configuration information indicating the recommended memory configuration to an output destination.