Operating system level resource management for real-time systems

The main goal of a real-time operating system (RTOS) is to provide foundations for guaranteeing deterministic or probabilistic timing requirements of real-time systems. In that regard, scheduling the CPU time has been the core aspect of any RTOS, as it directly contributes to satisfying timing requirements of real-time systems. In recent years, however, the emergence of multicore processor architecture and the growing number of battery powered devices make other aspects---such as contention in concurrently shared hardware resources and power consumption---also critical in providing required real-time performance. Management of concurrently shared hardware resources---such as shared caches, DRAM controllers, DRAM modules, hardware prefetchers---is important because contention in the shared resources can significantly degrade applications' execution times. Power (energy) management is also important because power-saving techniques typically have significant performance implications. This dissertation describes new OS level mechanisms and policies that control the shared hardware resources in commodity processors in ways that improve performance isolation and reduce energy consumption of real-time systems. These techniques enhance a RTOS's resource management capability so that it can be flexibly tailored to meet challenging real-time requirements without significantly compromising overall performance or wasting energy. Contributions of this dissertation include the following: (1) The design, prototype implementation, and performance evaluation of an efficient fine-grained memory bandwidth reservation system called MemGuard that improves performance isolation of multicore based real-time systems. (2) The design, model validation, and performance evaluation of the MultiDVFS scheme that jointly optimizes CPU and memory frequencies/voltages for real-time systems