From Tracepoints to Timeliness: A Semi-Markov Framework for Predictive Runtime Analysis

TL;DR

This paper introduces a semi-Markov framework that combines runtime tracing and probabilistic modeling to accurately predict worst-case execution times in real-time systems with minimal data and overhead.

Contribution

It presents a novel hybrid approach using semi-Markov chains for predictive runtime analysis, improving accuracy and efficiency over traditional measurement-based methods.

Findings

Accurately captures regular and extreme timing behaviors

Requires fewer assumptions and less data

Demonstrates effectiveness on real-time Linux systems

Abstract

Detecting and resolving violations of temporal constraints in real-time systems is both, time-consuming and resource-intensive, particularly in complex software environments. Measurement-based approaches are widely used during development, but often are unable to deliver reliable predictions with limited data. This paper presents a hybrid method for worst-case execution time estimation, combining lightweight runtime tracing with probabilistic modelling. Timestamped system events are used to construct a semi-Markov chain, where transitions represent empirically observed timing between events. Execution duration is interpreted as time-to-absorption in the semi-Markov chain, enabling worst-case execution time estimation with fewer assumptions and reduced overhead. Empirical results from real-time Linux systems indicate that the method captures both regular and extreme timing behaviours accurately, even from short observation periods. The model supports holistic, low-intrusion analysis across system layers and remains interpretable and adaptable for practical use.