Srinidhi Srinivasan

Real-time systems are systems in which correctness depends not only on producing the correct result but also on producing it within the required timing bounds. In many modern systems, including safety-critical and business-critical applications, violation of timing constraints can lead to unacceptable outcomes even when the underlying computation is correct. Ensuring that such systems consistently satisfy their timing requirements before deployment is therefore a fundamental challenge in the design and analysis of real-time systems.

Analysing the timing behaviour of these systems is challenging because their execution can vary across many possible scenarios. Variability in factors such as task release times, execution durations, precedence constraints among tasks, suspension intervals while tasks wait for remote computation or communication, can lead to a large number of possible execution scenarios, each potentially producing a different schedule and resulting in different timing properties. As systems become more complex, the number of such scenarios grows rapidly, making it difficult to reason about all possible outcomes in a systematic way.

The primary objective of this thesis is to improve the analysis of timing behaviour of realistic real-time systems. Modern systems often involve more complex task behaviour, where tasks may consist of multiple parts with precedence constraints or may temporarily pause while waiting for external operations. They also increasingly rely on communication networks to exchange data between different parts of the system. Together with the growing scale of such systems, these factors complicate the analysis of timing constraints. This thesis therefore proposes methods that enable systematic timing analysis across several challenging settings.

The first contribution of this thesis is related to the analysis of real-time systems that exhibit more expressive application-level task behaviour. In many practical systems, tasks cannot be described solely by simple recurring patterns: their activations may depend on events, interactions with other tasks, or delays between successive executions. Such behaviour increases the number and diversity of possible execution scenarios, making systematic timing analysis more difficult. This thesis develops methods to analyse the timing behaviour of real-time systems in which tasks may contain multiple parts with precedence constraints and may suspend while waiting for external operations. The proposed techniques model these precedence constraints and suspension intervals explicitly, and systematically reason about the execution scenarios that arise from such task behaviour.

The second contribution of the thesis concerns the analysis of timing behaviour in communication networks based on Time-Sensitive Networking (TSN). In these systems, the entities to be scheduled are network messages rather than processor jobs, and timing guarantees must be provided for packet transmissions across communication links. Analysing message latency in such networks is challenging because transmissions interact across multiple links, traffic shaping mechanisms regulate how packets are released and transmitted, and the internal structure of network switches influences how packets are queued and forwarded. These factors introduce complex dependencies that affect end-to-end communication delays. This thesis therefore investigates how communication timing properties in TSN networks can be analysed systematically. These methods provide analysis techniques for reasoning about how different traffic shaping mechanisms affect message latency, and support the analysis of both latency at individual network switches and end-to-end latency across communication paths.

The third contribution relates to the scalability of timing analysis for increasingly complex real-time systems. As the number of tasks, sources of timing variability, and interactions between system components grow, the number of execution scenarios that must be considered can increase significantly. This makes it difficult to analyse large systems while maintaining complete coverage of relevant scheduling possibilities. The thesis therefore investigates techniques that enable scalable timing analysis while preserving correctness. These techniques reduce the number of execution scenarios that must be explored while still ensuring complete coverage of relevant scheduling behaviours.

To enable these analyses, this thesis builds upon and enhances the Schedule Abstraction Graph (SAG) framework, originally introduced in [46, 48], which provides an abstraction-based approach for reasoning about scheduling behaviour. The core idea of the SAG is to represent the system’s possible schedules as a graph which captures relevant execution scenarios without analysing each schedule individually. Instead of representing every concrete execution scenario, the framework groups scenarios with equivalent scheduling behaviour into abstract states and explores their transitions. This enables the systematic analysis of timing behaviour across a wide range of execution scenarios.

Using this abstraction-based approach, the thesis extends the analysis to task models with precedence constraints and suspension behaviour, adapts the framework to analyse message transmissions in Time-Sensitive Networking, and develops techniques that improve the scalability of the analysis for larger systems. Taken together, these contributions improve the ability to analyse the timing behaviour of realistic real-time systems across several challenging settings.