| The increased processing power of computer systems allows new applications to emerge, including applications with real-time requirements. For example, applications for multimedia have become ubiquitous during the last decade. Traditionally, real-time applications execute on dedicated hardware platforms or use special-purpose operating systems; now real-time workloads exist in general-purpose platforms that are ill-equipped to meet their time constraints. To support applications with real-time requirements, while still servicing the existing workload in a time-sharing discipline, general-purpose operating systems must be capable of concurrently scheduling a variety of different timing constraints.; General-purpose time-sharing schedulers do not support either the predictability or timeliness requirements of real-time tasks, and it is difficult to add real-time capability to a system that does not already support real-time scheduling. In contrast, real-time schedulers provide predictability and timeliness, and it is easier to schedule non-real-time tasks in a real-time system. Therefore, this thesis argues that to support the wide variety of existing and future application constraints, any "general-purpose" system is best served by using a real-time CPU scheduler. This dissertation describes a system that uses a real-time CPU scheduler to schedule non-real-time applications in the manner of a time-share discipline. |
