A dynamic CPU bandwidth partitioning scheme for multimedia real-time systems

Advances of computer and network technology have expanded multimedia applications to very broad ranges. An operating system that supports not only generic best effort applications but also multimedia real-time applications is required. But there is no operating system to support various types of applications in a system. The proposed algorithm supports several scheduling methods under an operating system. The algorithm groups applications into several classes, partitions CPU bandwidth, and assigns a portion of CPU bandwidth according to the class of applications. Depending upon the applications, the algorithm allocates a proper scheduler for each application. And a two-level scheduler is deployed to schedule each class and task. The first-level scheduler selects a class by the rate and passes a time quantum to the second-level scheduler. The second-level scheduler schedules tasks by a conventional scheduling algorithm such as RR(round-robin) or EDF. Under overloaded condition, the algorithm does not affect other classes of applications. By using static rate for an application class, the algorithm not only prevents starvation for best effort applications but also guarantees the minimal execution of applications.