The Case for Hierarchical Schedulers with Performance Guarantees

Audio and video applications, process control, agile manufacturing and even defense systems are using commodity hardware and operating systems to run combinations of real-time and non-real-time tasks. We propose an architecture that will allow a general-purpose operating system to schedule conventional and real-time tasks with diverse requirements, to provide flexible load isolation between applications, users, and accounting domains, and to enforce high-level policies about the allocation of CPU time. This is accomplished by implementing a dynamic, hierarchical scheduling infrastructure. The infrastructure is integrated with a resource manager that provides a level of indirection between resource requests and the scheduling hierarchy. A scheduling infrastructure separates scheduler code from the rest of the operating system. To demonstrate the utility of our architecture, we describe its application to three existing real-time schedulers. For each of the three, we show added flexibility while retaining the original scheduling guarantees.