Search Results

Hard real time systems are evolving in order to respond to the increasing demand in complex functionalities while taking advantage of newer hardware. Software development for safety critical systems has to comply with strict requirements that will facilitate the certification process. During this process, each part of the system is evaluated, requiring a certain level of assurance in order to provide confidence in the product. In particular there must be a level of confidence that the system behaves deterministically that may be based on functionality, resources and time. The success of system verification depends greatly on the capacity to determine its exact behavior. Nonetheless, hardware evolved in order to maximize the average computation power throughput with little to no regard to the deterministic aspect. Therefore modern architectural features of processors, like pipelines, cache memories and co-processors, make it hard to verify that all the needed properties are respected. The multi-core is furthermore difficult to analyze as the architecture employs mechanisms that compromise strong spatial and temporal partitioning when using shared resources without rigorous access control like shared caches or shared input/outputs. In this paper we identify and analyze the main sources of nondeterminism of the multi-cores with regard to the timing estimation. Precise determination of the worst case execution time is a challenging task even in single-core architectures. The problems are accentuated in the multi-core context mainly due to the resource sharing that can lead to highly complex interactions or to nondeterminism. Most of the units that generate behaviors that are hard to take into account can be deactivated, but it is not always easy to predict the impact on the performance. Nevertheless some of the features cannot be disabled (such as the out of order execution or some nondeterministic crossbar access policies) which leads to the invalidation of the respective platform for applications with high criticality level. We will address the problematic units, propose configuration or architecture guidelines and estimate their impact on the performance and determinism of the system.