A distributed open environment for real-time applications.

摘要

Traditional approaches to real-time schedulability analysis tend to require detailed timing attributes and resource usages of all applications that may run concurrently in the system. Reconfiguring such a system is expensive and often done offline. This fact has motivated the design of open system architecture. Open system is a hierarchical reservation-based framework that provides temporal guarantee and isolation to independently developed applications. It is now possible to decide at runtime whether an application can meet its timing requirement based on a few overall parameters. However, open system was originally developed for applications to run on a single processor. This dissertation generalizes the technique and extends the open architecture to a distributed environment. Many real-time applications are distributed in nature and require reservation support at the end-hosts as well as within the network. In a distributed system where a hybrid set of applications (i.e. hard, soft, and non real-time applications) may run concurrently in a cluster of processors connected together by a local area network, "temporal isolation" becomes a highly desirable property. The proposed distributed open architecture manages the CPU and network resources, and provides end-to-end timing guarantee to all real-time applications that are admitted into the system. The developer of each real-time application validates the schedulability of the application assuming it is running alone on a dedicated platform. If a real-time application is schedulable on a slower system of the required capacity and the open system admits the application, its temporal behavior will not be affected by any other applications in the system. This dissertation validated the distributed open system for different types of networks: Myrinet, Controller Area Network (CAN), and Wireless LAN. While the hierarchical scheduling framework remains the same, several scheduling algorithms are developed to address the design issues arise from the hardware constraints and/or industrial standards of each network. Simulation results show that real-time guarantee and inter-application isolation can be achieved with acceptable scheduling overhead. In some cases, the hierarchical scheduling scheme reduces the problem complexity and increases network utilization.