Graph Computing Performance Prediction Using Multi-Feature Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for performance prediction in graph computing tasks are inaccurate and lack comprehensive analysis, particularly due to the large volume and complexity of graph data, making it difficult to estimate execution efficiency effectively.

Innovation Solution

A performance prediction model is developed that utilizes training samples from various graph processing platforms, incorporating features like graph data, processing platform, algorithm, and hardware characteristics to predict execution times, allowing for optimal environment determination and improved accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual analysis methods are used to estimate graph computing task performance, then the analysis process is simple and intuitive, but the prediction accuracy and comprehensiveness are insufficient

Engineering Contradiction:
Improveperformance prediction accuracyVSAvoidmodel complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The performance prediction model segments the graph computing task into multiple independent feature dimensions: graph data features (number of vertices, edges, density), processing platform features (processor type, memory capacity, storage capacity), algorithm features (algorithm type, computational complexity), and hardware features (GPU model, CPU model). Each dimension is evaluated separately and then integrated to provide comprehensive performance prediction, resolving the contradiction between accuracy and complexity by making the complex model structured and manageable.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms the performance prediction approach by changing from manual qualitative analysis to automated quantitative parameter-based prediction. Multiple measurable parameters are collected and processed through machine learning models to predict execution time, resource utilization, and cost metrics objectively, significantly improving prediction accuracy while maintaining model interpretability through standardized parameter definitions.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If comprehensive feature analysis is performed to improve prediction accuracy, then the prediction result becomes more accurate, but the computational cost and time consumption increase

Engineering Contradiction:
Improveperformance prediction accuracyVSAvoidmodel training and prediction time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary actions by pre-collecting and storing graph data features, platform features, algorithm features, and hardware specifications in structured formats before actual performance prediction is needed. Feature extraction and normalization are performed in advance, and the trained model can quickly make predictions without reprocessing raw data, significantly reducing the time required for actual performance estimation while maintaining comprehensive analysis.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If multiple performance indicator features are incorporated into the prediction model, then the comprehensiveness of analysis improves, but the complexity of model configuration and usage increases

Engineering Contradiction:
Improveanalysis comprehensivenessVSAvoidmodel configuration ease
Core Design Contradiction:
Adaptability or versatilityVSEase of operation

Solution Approach 1:

The performance prediction model is designed with universal applicability across different graph computing platforms and algorithms. The same model structure and feature extraction methodology can be applied to various graph databases (Neo4j, TigerGraph, Amazon Neptune), different processors (CPU, GPU, FPGA), and multiple graph algorithms (PageRank, Connected Components, Shortest Path). This multi-functionality allows comprehensive analysis without requiring separate complex configurations for each scenario, as the model adapts to different platforms through standardized feature inputs.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3955204B1Data processing method and apparatus, electronic device and storage medium
Publication Date: 2025.12.03 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3955204B1 patent drawingFigure 1
  • EP3955204B1 patent drawingFigure 2
  • EP3955204B1 patent drawingFigure 3

AI summary

Disclosed are a data processing method and apparatus, an electronic device and a storage medium. The method comprises: acquiring a training sample set, wherein the training sample set comprises a plurality of graph calculation task training samples; extracting performance index features corresponding to the graph calculation task training samples; generating a target performance prediction model on the basis of a mapping relationship between actual operation times of the graph calculation task training samples and the corresponding performance index features; and on the basis of the target performance prediction model, outputting, according to a performance index feature corresponding to an input graph calculation task test sample, a predicted operation time of the graph calculation task test sample.