Measuring performance of soft real-time tasks on multi-core systems

To take advantage of multicore platforms in embedded systems, soft real-time applications are run together with general purpose applications. Due to shared hardware resources in multicore architectures, it is likely that these applications will interfere and compete with each other. In order to investigate this interference, real-time tasks were co-scheduled with general purpose tasks. For performance measurement a test environment was created that uses hardware performance counters to count core, level-2 cache and memory bus events at run-time. As a result of our study we can say that co-existence is possible without major performance degradation, if the general purpose task is not much memory intensive. Otherwise there is a trade-off between the number of cores executing general purpose applications and the amount of tolerance an embedded system can have in response times.