TMLL Trace Server Machine Learning Library, Use AI for Trace Analysis
0Presented by Kaveh Shahedi and Matthew Khouzam virtually at TheiaCon 2024.
In performance analysis, tracing is essential for capturing detailed execution data and enabling precise analysis of system behavior and resource usage. Trace Compass, an open-source tool developed at Ericsson, provides extensive views, graphs, and metrics for system trace analysis. However, using it effectively often requires prior knowledge of specific analyses. In this work, we propose a fully automated pipeline that simplifies the use of Trace-Server, an independent standalone derived from Trace Compass, by applying machine learning to gain valuable insights from trace outputs. This tool allows users to analyze their system traces seamlessly within Theia, offering a more user-friendly and streamlined experience. Additionally, Jupyter Notebook-based analyses are integrated to make the application of machine learning and statistical techniques more interactive and intuitive for trace analysis.