Guaranteeing end-to-end deadlines in distributed client-server applications

Abstract

The paper presents a scheme for guaranteeing the scheduling of real-time computations in a distributed environment. The authors propose a process model where only true (local or end-to-end) deadlines need to be specified. They assume all local interactions among processes are based on shared memory communication, protected by priority ceiling semaphores. Remote interactions are client-server blocking communications. The result is both a scheduling policy and a methodology to guarantee the integrated scheduling of processes and network messages. Both processes and messages are scheduled according to a fixed-priority scheme that can easily be implemented on most operating systems and a few network protocols. The procedure that checks the schedulability of the distributed computation against end-to-end deadlines is simple enough to be proposed as a scheme for a dynamic guarantee.