Deriving Efficient and Dependable Parallel Programs from Simulink Models

Modern highly-demanding applications require high computational power, on the one hand,and fulfilment of real-time constraints and a high level of resilience, on the other. Model-based development provides means to address these objectives. Simulink is one such widely used tool for model-based development of software. It supports a complete design chain starting from modelling and ending in code generation. However, the programs generated by the built-in code generator cannot fully utilize the potential offered by energy and performance efficient many-core platforms that are likely to be used in various application domains: from home use electronics to complex critical control systems. Furthermore, the failure of a processing core where the derived implementation runs leadsto the failure of the application. In contrast, the distribution of tasks among processing cores allows for invoking dynamic reconfiguration procedures, e.g., tasks reallocation, so that the required level of resilience is achieved and efficiency is maintained at an acceptable level. Despite this, however, the application tasks may lose data when they are dynamically reallocated to mask failures of individual cores. This paper addresses the problems described above by proposing: 1) an approach to generation of parallel implementations from Simulink models, 2) relying on 1), a mechanism that prevents data loss when application tasks are reallocated. The paper also presents evaluation results of the proposed approach on an industrial case study using a commercially available NoC-based platform, namely TilePro by Tilera. The evaluation shows only about 1% performance degradation when comparing an FT implementation with a non-FT one.