Bounding Resource Contention Interference in the Next-Generation Microprocessor (NGMP)

Abstract

The Space industry, as several other real-time industries , is assessing the use of multicore processors as their main computing platform. While multicore processors bring the potential of integrating several software (mixed-criticality) functions , their use also brings some challenges. In particular, tasks running in multicores may experience high contention delays when accessing multicores' shared resources. This makes that the load that a task puts on shared resources impacts the Execution Time Bounds (ETBs) derived for other corunning tasks. In this paper we focus on the Cobham Gaisler NGMP – acknowledged as one of the multicore processors currently assessed by the European Space Agency for its future missions – for which we propose a measurement-based approach to bound contention interference. Given a task τ , instead of providing ETBs for the highest contention that any set of corunners can generate – already shown to be potentially high – our approach provides bounds that factor in the number of requests contenders generate regardless of how they align with τ 's requests. This provides a good balance between ETBs accuracy and independence from the corunners, since our approach only requires controlling the number of requests each task makes to the shared resources.