Design And Implementation Of Semi-partitioned Scheduler With Shared Resource Management

Real-time systems exist in many areas, such as avionics, automotive electronics. To satisfy the increasing requirements on higher performance, it has been widely believed that future real-time systems will adopt multicore processors.Real-time scheduling has always been a hot research issue in the field of real-time system. Compared with traditional single-core systems, multicore systems usually have on-chip shared resources. Real-time tasks on different cores access the shared resources, which makes the timing of the real-time system unpredictable. Resource isolation is an effective principle to reduce the interference between tasks. This technique can also be used to improve the timing predictability of multicore real-time systems.In this thesis, we designed and implemented a semi-partitioned multicore real-time scheduler. Meanwhile, we investigated combining spatial ioslation techniques with multicore real-time scheduling policies. Specifically, a semi-partitioned scheduler is implemented based on the Linux kernel. Then the related overheads are measured, such as migration of the task in semi-partitioned scheduler. The result shows that the related overheads of the scheduler are small with little effect on the semi-partitioned scheduler. Second, page coloring based shared cache isolation is extended in our platform. In this system platform, experiments are conducted to test the fluctuations of the execution time of real-time tasks. The results show that, for the shared cache conflict sensitive real-time tasks, the fluctuations of real-time task execution time are significantly reduced by cache isolation, and the timing predictablity is improved. Therefore, deadline misses of the real-time tasks can be reduced.