Methodology of ML in Performance Testing (How to Find Bottlenecks Automatically)
summary
The talk describes an ML-based performance-testing pipeline that processes nightly load-test logs in MSSQL and a few hundred server metrics. It uses R and k-means clustering to group resource-usage patterns, ARIMA-style time-series forecasting to predict medians and percentiles, and a correlation matrix plus diagnostic micro-requests and propagated trace headers to map bottlenecks to service layers; heuristic thresholds and reproducibility across runs determine whether alerts are filed, with HTML/XLS reports generated for developers.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:57Automated data analysis need
- 7:09Log categories
- 10:48Diagnostic queries per layer
- 15:25ML for expected results
- 17:16Server monitoring limits
- 23:10Clustering anomalies
- 24:14Correlation analysis
- 27:26Request tracing
- 30:43Regression analysis
- 33:36Visualizing results
- 37:01Data storage processing
related talks
14mPerformance Optimization in High-Volume Transaction Systems
29mEnhancing Performance Test Results with ML & LLMs: Communication Beyond the Numbers
10mML-based Log Analysis for Faster Debugging
14mLeveraging ML to Detect Application HotSpots [@scale, of Course!]
41mUsing ML to find value in your automated tests
44mML in Production
25mChallenges and Experiences with MLOps for Performance Diagnostics in Hybrid-Cloud
28mML-Driven Test Intelligence at Scale - What Works What Fails and Why It Matters