Time-predictable execution of multithreaded applications on multicore systems

In multicore systems, contention for access to main memory between application threads complicates timing analysis and may lead to pessimistic bounds on execution time. This is particularly problematic for real-time applications, which require provable bounds on worst-case performance. In this work, we employ a predictable execution model to schedule memory accesses performed by application threads without relying on unpredictable hardware arbiters. In addition, we statically schedule application´s threads with the objective to minimize the application´s makespan. Our experimental evaluation with NAS Parallel Benchmarks on 4-core system indicates that the proposed execution scheme yields an aggregated improvement of 21% over contention execution in which application´s threads uncontrollably access main memory.