Efficient and Predictable Approach for Supporting Spatial Partitioning of Real-Time Applications

M. Younis and S. Ponnusamy

Keywords

Spatial portioning, realtime systems, memory management, integrated critical control

Abstract

A typical mission-critical real-time system is made up of a number of subsystems (applications), each allocated separate hardware and software resources. However, integration of multiple real-time applications has gained acceptance as a new trend in the industry during the past few years. Integration would require real-time applications to share computing resources and a significant savings in cost and space can be achieved. However, sharing of computing resources raises the issue of spatial partitioning between real-time applications in order to prevent a subsystem-level error from causing a system-wide failure. The provision of spatial partitioning in most advanced processors involves the use of virtual addressing. Virtual addressing is rarely used in real-time systems as it introduces unpredictability to the execution time and thus causes overestimation of the computing and communication resources required to guarantee timeliness. In addition, address translation introduces overhead that slows down execution. In this paper, we describe an efficient and yet predictable approach for supporting spatial partitioning in mission-critical real-time applications. Our approach classifies real-time tasks based on criticality and communication patterns and tailors the implementation of spatial protection to the need of tasks. We show that our approach can provide a tight estimate for the execution time and allows efficient resource utilization. The approach is validated on the PowerPC architecture.

Important Links:



Go Back