“Real-Time Scheduling in Multicore Time- and Space-Partitioned Architectures”
Revision as of 12:56, 25 July 2014 by Jcraveiro
Ph.D. dissertation, University of Lisbon, Lisbon, Portugal, Aug. 2013Defended on June 2, 2014. Approved with distinction and praise, by unanimity
Abstract: The evolution of computing systems to address size, weight and power consumption (SWaP) has led to the trend of integrating functions (otherwise provided by separate systems) as subsystems of a single system. To cope with the added complexity of developing and validating such a system, these functions are maintained and analyzed as components with clear boundaries and interfaces. In the case of real-time systems, the adopted component-based approach should maintain the timeliness properties of the function inside each individual component, regardless of the remaining components. One approach to this issue is time and space partitioning (TSP) — enforcing strict separation between components in the time and space domains. This allows heterogeneous components (different real-time requirements, criticality, developed by different teams and/or with different technologies) to safely coexist. The concepts of TSP have been adopted in the civil aviation, aerospace, and (to some extent) automotive industries. These industries are also embracing multiprocessor (or multicore) platforms, either with identical or non-identical processors, but are not taking full advantage thereof because of a lack of support in terms of verification and certification. Furthermore, due to the use of the TSP in those domains, compatibility between TSP and multiprocessor is highly desired. This is not the present case, as the reference TSP-related specifications in the aforementioned industries show limited support to multiprocessor. In this dissertation, we defend that the active exploitation of multiple (possibly non-identical) processor cores can augment the processing capacity of the time- and space-partitioned (TSP) systems, while maintaining a compromise with size, weight and power consumption (SWaP), and open room for supporting self-adaptive behavior. To allow applying our results to a more general class of systems, we analyze TSP systems as a special case of hierarchical scheduling and adopt a compositional analysis methodology.
Research line(s): Timeliness and Adaptation in Dependable Systems (TADS)