Strict real-time systems design over multiprocessor-based platforms

• Autores: Jorge Garrido Balaguer
• Directores de la Tesis: Juan Antonio de la Puente Alfaro (dir. tes.)
• Lectura: En la Universidad Politécnica de Madrid ( España ) en 2018
• Idioma: español
• Tribunal Calificador de la Tesis: Alejandro Alonso Muñoz (presid.), Alfons Crespo i Lorente (secret.), Michael González Harbour (voc.), Tullio Vardanega (voc.), Luis Miguel Rosario da Silva Pinho (voc.)
• Programa de doctorado: Programa de Doctorado en Ingeniería de Sistemas Telemáticos por la Universidad Politécnica de Madrid
• Materias:
  • Matemáticas
    • Ciencia de los ordenadores
      • Calculo digital
    • Investigación operativa
      • Sistemas de control
  • Ciencias tecnológicas
    • Tecnología de los ordenadores
      • Arquitectura de ordenadores
      • Sistemas en tiempo real
• Enlaces
  • Tesis en acceso abierto en: Archivo Digital UPM
• Resumen

  • In the past decades the digitalization and automation have played a key role in the economic growth and development. An increasing number of activities supporting the levels of welfare achieved are based on the adoption of modern technologies in which computers have enabled higher service levels: communications, transportation, industrial production, health care or education are some examples. Nowadays energy production and distribution are controlled by computer systems, as well as our daily communications via mobile phones and emails, and we rely on many electronic-driven safety measures now standard in the automotive industry.

    Most of these embedded computer systems are in charge of controlling complex critical systems concerning human lives and business viability. These systems present strict requirements to ensure proper behaviour. Some of these requirements address non-functional aspects, as the temporal correctness of the system. In other words, for some relevant computer-controlled systems it is not only important to produce the correct output but also to produce it on time.

    This dissertation aims to contribute on the development of systems with strict temporal requirements based on multiprocessor platform. Modern hardware technology has enabled the massive production of processors incorporating more than one computation unit per chip, increasing the overall computer performance. These new platforms present, however, a set of challenges to achieve the required reliability according to the critical environments in which they operate.

    In this dissertation the temporal analysis of real-time systems based on multiprocessor platforms is addressed. In particular it contributes in a number of ways on the definition and study of a new scheduling protocol, the Multiprocessor resource sharing Protocol (MrsP). This protocol aims to transfer the most successful approaches from mono processor theory and practice in combination with the most promising techniques for strict multiprocessor real-time systems design. This is substantiated in the use of ceiling priority and equivalent stack re- source policies as task dispatching strategies with a preemptable spin-locking mechanism for shared resource access arbitration. Global spin-delay interference is mitigated by a novel helping mechanism, based on the cooperative completion of resource accesses locally preempted.

    Specific contributions addressed in this work include the formalization of a more exact timing analysis based on the heterogeneity of access times present in practice. The task model is then completed with support for nested resource access patterns, including means for analysing this task model with two scheduling analysis based on the well known response time analysis technique. These two different analysis can prove the system feasibility with different degrees of system knowledge: while the sufficient analysis enables testing the schedulability of the system with reduced knowledge of resource access patterns, a tighter, less pessimistic analysis can achieved when all shared resource access patterns have been fully characterized.

    These contributions are supported by the extensive evaluation work already published. The evaluation section of this dissertation complements that work with the analysis of the flight segment of a space mission which is then compared with that of a traditional monoprocessor approach as well as with a comparable multiprocessor protocol.