Dependable Real-Time Systems
Abstract
Most current real-time operating systems contain the same basic paradigms found in timesharing operating systems and are simply stripped down and optimized versions of timesharing operating systems. For example, while they stress fast mechanisms such as a fast context switch and the ability to respond to external interrupts quickly, they retain the main abstractions of timesharing operating systems. In addition, very often today's real-time kernels use priority scheduling. Priority scheduling is a mechanism which provides no direct support for meeting timing constraints. For example, the current technology burdens the designer with unenviable task of mapping a set of specific constraints on task executions into task priorities in such a manner that all tasks will meet their deadlines. Thus, when using the current paradigms together with priority scheduling it is difficult to predict how tasks, dynamically invoked, interact with other active tasks, where blocking over resources will occur, and what the subsequent effect of this interaction and blocking is on the timing constraints of all the tasks. Basically, currently used scheduling policies are inadequate for three main reasons: (1) they do not address the need for an integrated central processing unit scheduling and resource allocation scheme, (2) they don't handle the end-to-end scheduling problem, and (3) they are not used in a planning mode, thereby, containing a myopic view of the system capabilities.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 30, 1991
- Accession Number
- ADA271770
Entities
People
- John A. Stankovic
- Krithi Ramamritham
- Victor Yodaiken
Organizations
- University of Massachusetts Amherst