Scheduling and Analysis of Real-Time OpenMP Task Systems with Tied Tasks

OpenMP is a promising programming framework to develop parallel real-time systems on multi-cores. Although similar to the DAG task model, the OpenMP task systems are significantly more difficult to analyze due to various constraints posed by the OpenMP specification. An important feature in OpenMP is the tied tasks, which must execute on the same thread during the whole life cycle. Although tied tasks enjoy benefits in simplicity and efficiency, it was considered to be not suitable to real-time systems due to its complex behavior. In this paper, we study the real-time scheduling and analysis of OpenMP task systems with tied tasks. First, we show that under the existing scheduling algorithms adopted by OpenMP, tied tasks indeed may lead to extremely bad timing behaviors where the workload of a parallel task system is sequentially executed. To solve this problem, we propose a new scheduling algorithm and we developed two response time bounds for the new algorithm, with different trade-off between simplicity and analysis precision. Experiments with both randomly generated OpenMP task systems and realistic OpenMP programs show that the response time bounds obtained by our new scheduling algorithm and analysis techniques for tied task systems are very close to that of untied tasks, which may also be a good choice for real-time systems in many cases.