This thesis addresses the problem of online multi-resource management in embedded real-time systems. It focuses on three research questions. The first question concentrates on how to design an efficient hierarchical scheduling framework for supporting independent development and analysis of component based systems, to provide temporal isolation between components. The second question investigates how to change the mapping of resources to tasks and components during run-time efficiently and predictably, and how to analyze the latency of such a system mode change in systems comprised of several scalable components. The third question deals with the scheduling and analysis of a set of parallel-tasks with real-time constraints which require simultaneous access to several different resources. For providing temporal isolation we chose a reservation-based approach. We first focused on processor reservations, where timed events play an important role. Common examples are task deadlines, periodic release of tasks, budget replenishment and budget depletion. Efficient timer management is therefore essential. We investigated the overheads in traditional timer management techniques and presented a mechanism called Relative Timed Event Queues (RELTEQ), which provides an expressive set of primitives at a low processor and memory overhead. We then leveraged RELTEQ to create an efficient, modular and extensible design for enhancing a real-time operating system with periodic tasks, polling, idling periodic and deferrable servers, and a two-level fixed priority Hierarchical Scheduling Framework (HSF). The HSF design provides temporal isolation and supports independent development of components by separating the global and local scheduling, and allowing each server to define a dedicated scheduler. Furthermore, the design addresses the system overheads inherent to an HSF and prevents undesirable interference between components. It limits the interference of inactive servers on the system level by means of wakeup events and a combination of inactive server queues with a stopwatch queue. Our implementation is modular and requires only a few modifications of the underlying operating system. We then investigated scalable components operating in a memory-constrained system. We first showed how to reduce the memory requirements in a streaming multimedia application, based on a particular priority assignment of the different components along the processing chain. Then we investigated adapting the resource provisions to tasks during runtime, referred to as mode changes. We presented a novel mode change protocol called Swift Mode Changes, which relies on Fixed Priority with Deferred preemption Scheduling to reduce the mode change latency bound compared to existing protocols based on Fixed Priority Preemptive Scheduling. We then presented a new partitioned parallel-task scheduling algorithm called Parallel-SRP (PSRP), which generalizes MSRP for multiprocessors, and the corresponding schedulability analysis for the problem of multi-resource scheduling of parallel tasks with real-time constraints. We showed that the algorithm is deadlock-free, derived a maximum bound on blocking, and used this bound as a basis for a schedulability test. We then demonstrated how PSRP can exploit the inherent parallelism of a platform comprised of multiple heterogeneous resources. Finally, we presented Grasp, which is a visualization toolset aiming to provide insight into the behavior of complex real-time systems. Its flexible plugin infrastructure allows for easy extension with custom visualization and analysis techniques for automatic trace verification. Its capabilities include the visualization of hierarchical multiprocessor systems, including partitioned and global multiprocessor scheduling with migrating tasks and jobs, communication between jobs via shared memory and message passing, and hierarchical scheduling in combination with multiprocessor scheduling. For tracing distributed systems with asynchronous local clocks Grasp also supports the synchronization of traces from different processors during the visualization and analysis.
|Kwalificatie||Doctor in de Filosofie|
|Datum van toekenning||16 okt 2012|
|Plaats van publicatie||Eindhoven|
|Status||Gepubliceerd - 2012|