Encoder-Decoder Model for Workload Performance Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Assessing the specific contribution of hardware components on workload performance is challenging due to the complex interplay of various components, making it difficult to predict performance benefits across different hardware platforms without laborious source code instrumentation or inaccurate correlation of time-series execution data.
Innovation Solution
An encoder-decoder machine learning model is trained using time-series execution performance information from both source and target hardware platforms, creating a mapping to predict workload performance on a target platform relative to a known source platform without requiring time-interval segmentation or source code instrumentation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional source code instrumentation or time-series correlation methods are used to assess hardware component contribution, then performance prediction can be achieved, but the process becomes laborious and complex
Solution Approach 1:
The patent replaces traditional mechanical/manual source code instrumentation and complex time-series correlation analysis with an AI-based machine learning system. The AI model automatically analyzes hardware platform characteristics and workload performance data to predict performance benefits, eliminating the need for manual code modification and complex mathematical correlations while maintaining high prediction accuracy
Solution Approach 2:
The patent introduces an AI-based intermediary system that mediates between hardware platform specifications and performance prediction outcomes. This intermediary automatically processes hardware component information, executes workloads, collects performance metrics, and generates predictions without requiring direct manual analysis or source code modification, thereby simplifying the overall assessment process
2Measurement precision
If detailed time-series execution data correlation is performed, then performance assessment can be achieved, but the method becomes inaccurate and laborious
Solution Approach 1:
The patent performs preliminary actions by pre-training the AI model on diverse hardware platforms and workload combinations before actual performance assessment. The model learns to recognize patterns and relationships between hardware characteristics and performance outcomes in advance, enabling rapid and accurate predictions without requiring laborious real-time data processing or manual correlation analysis
Solution Approach 2:
The patent substitutes manual time-series data collection, alignment, and correlation processes with automated AI-based analysis. The system automatically executes workloads on different hardware platforms, collects performance metrics, and uses machine learning to correlate hardware characteristics with performance outcomes, dramatically reducing the time required while improving accuracy through pattern recognition
Data Source
AI summary
For each of a number of workloads, first time-series execution performance information is collected during execution of the workload on a first hardware platform. For each workload, second time-series execution performance information is collected during execution of the workload on a second hardware platform. An encoder-decoder machine learning model is trained that outputs predicted performance on the second hardware platform relative to known performance on the first hardware platform. The encoder-decoder machine learning model is trained from the first and second time-series execution performance information for each workload.