The Performance Enhancement of Real Time Embedded Applications Mapping to Multicore Systems

Abstract

In this thesis, we presented a new kernel based solution for overload in mixed criticality multicore systems. The solution is based on virtual timer, which is used to control release times and priority points of level-C jobs under the overload state. This virtual timer does not affect high level tasks. Moreover, it has no effect on execution time. The virtual timer is used immediately after detecting an overload state to slow down the releases of coming level-C tasks for limited time until the system recover from overload. The system returns back to normal time after having all tasks’ response times back to normal. We focused on the fast response to overload state. This is achieved by implementing the overload handling inside the kernel space. Despite that KSO inherits the main idea of using virtual timer to recover from overload from GEL-v, KSO has a completely different architecture and implementation. Unlike GEL-v, our proposed technique KSO avoids using any user space invocation as it implements the overload handling completely inside the kernel space. KSO eliminated many steps in implementation and removed causes of delays found in GEL-v. We applied many experiments to test the response time for overload in KSO. We compared between KSO response time to overload and GEL-v response time to overload. Our experimental results proved that KSO is faster in its response to overload. Also we showed that depending on user space has a bad effect on response to overload in addition to its overhead added to system tasks. 6.2 Contributions In this section, we summarize the most important contributions of this thesis as follows. • We gave an important background on the main challenges of scheduling on multicores especially when having mixed criticality systems. A detailed review of scheduling algorithms, used to map