Toward Contention Analysis for Parallel Executing Real-Time Tasks

In measurement-based probabilistic timing analysis, the execution conditions imposed to tasks as measurement scenarios, have a strong impact to the worst-case execution time estimates. The scenarios and their effects on the task execution behavior have to be deeply investigated. The aim has to be to identify and to guarantee the scenarios that lead to the maximum measurements, i.e. the worst-case scenarios, and use them to assure the worst-case execution time estimates. We propose a contention analysis in order to identify the worst contentions that a task can suffer from concurrent executions. The work focuses on the interferences on shared resources (cache memories and memory buses) from parallel executions in multi-core real-time systems. Our approach consists of searching for possible task contenders for parallel executions, modeling their contentiousness, and classifying the measurement scenarios accordingly. We identify the most contentious ones and their worst-case effects on task execution times. The measurement-based probabilistic timing analysis is then used to verify the analysis proposed, qualify the scenarios with contentiousness, and compare them. A parallel execution simulator for multi-core real-time system is developed and used for validating our framework. The framework applies heuristics and assumptions that simplify the system behavior. It represents a first step for developing a complete approach which would be able to guarantee the worst-case behavior