Software Telemetry Clustering for Performance Regression Diagnosis

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Identifying the specific software modifications associated with performance degradation in complex software systems is challenging due to the overwhelming number of telemetry counters and the lack of historical data, making it difficult to diagnose and fix performance regressions.

Innovation Solution

Utilizing machine learning self-organizing maps to cluster software telemetry counter time series data from original and modified software versions, enabling deviation analysis to identify discriminated counters that indicate performance differences.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If traditional manual analysis methods are used to identify software modifications causing performance degradation, then diagnostic accuracy can be maintained, but the time and effort required increases significantly

Engineering Contradiction:

Improvediagnosis timeVSAvoidanalysis system complexity

Core Design Contradiction:

Loss of timeVSDevice complexity

Solution Approach 1:

The patent introduces self-organizing maps as an intermediary computational mechanism between the raw telemetry counter data and the final diagnosis. These maps automatically cluster and organize the overwhelming number of counters, serving as a mediator that transforms complex data into actionable insights without requiring manual analysis of each counter, thereby reducing diagnosis time while managing complexity through automated pattern recognition

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces the mechanical manual analysis process with an automated machine learning system. Instead of manually examining telemetry counters and determining cause-and-effect relationships, the system uses self-organizing maps and automated deviation analysis to identify problematic modifications, significantly reducing the time and human effort required for diagnosis

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If all software telemetry counters are monitored and analyzed, then complete coverage of software behavior is achieved, but the complexity of analysis and data processing becomes overwhelming

Engineering Contradiction:

Improvesoftware behavior monitoring accuracyVSAvoiddata processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts and focuses on the most relevant information by using self-organizing maps to cluster telemetry counters based on their behavior patterns. Instead of analyzing all counters equally, the system extracts the key discriminatory features that actually indicate performance degradation, filtering out redundant information and reducing analysis complexity while maintaining measurement precision

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent segments the overwhelming set of telemetry counters into meaningful groups using self-organizing maps. By dividing the data into clusters based on similar behavior patterns, the system makes the analysis manageable and reduces complexity, allowing focused examination of specific counter groups rather than treating all counters as a monolithic dataset

Inventive Principle:

Principle #1Segmentation

3Extent of automation

If self-organizing maps are used to cluster telemetry counter time series, then automated detection of performance degradation causes is achieved, but computational resources and processing time for training increase

Engineering Contradiction:

Improveperformance degradation detection automationVSAvoidcomputational resource consumption

Core Design Contradiction:

Extent of automationVSUse of energy by moving object

Solution Approach 1:

The patent applies partial action by using the self-organizing maps only for the specific purpose of clustering and identifying discriminatory counters, rather than attempting to analyze every aspect of software behavior. This focused application achieves sufficient automation for performance degradation detection while avoiding the excessive computational resources that would be required for comprehensive analysis of all possible software metrics

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS12572454B2Identifying software modifications associated with software performance degradation

Publication Date: 2026.03.10 DELL PROD LP

AI summary

A software analysis system collects a first set of time series counter values from an original version of software instrumented with software telemetry counters, and collects a second set of time series counter values from a modified version of the software that also is instrumented with the software telemetry counters. The first time series become dimensions input to train a weight matrix of a first self-organizing map, to cause the first self-organizing map to cluster the time series into a group of first clusters describing execution of the original software. The second time series become dimensions input to train a weight matrix of a second self-organizing map, to cause the second self-organizing map to cluster the time series into a group of second clusters describing execution of the modified software. Deviation analysis between the first and second groups of clusters is used to identify execution differences between the software versions.