Eecient Methods of Validating Timing Constraints

Analytical and eecient validation methods to determine whether all jobs always complete by their deadlines are not yet available for systems using modern dynamic scheduling strategies. Exhaustive methods are often infeasible or unreliable since the execution time and release time of each job may vary. This chapter presents several worst-case bounds and eecient algorithms for determining how late the completion times of independent jobs with arbitrary release times can be in a dynamic multiprocessor or distributed system when their release times and execution times may vary from one instance to another. The special cases considered here are when the jobs are (1) preemptable and migratable, or (2) preemptable and nonmigratable, or (3) nonpreemptable. 9.1 Introduction In a real-time system, many jobs are time-critical. Here, by job, we mean a unit of work to be scheduled and executed. A job may be the computation of a control law, the transmission of an operator command, the retrieval of a le, etc. To execute, it requires a computer, a data link, a console, a disk, respectively; we refer to them all as processors. The processors are identical if they can be used interchangeably. Otherwise they are functionally dedicated, as exempliied by the processors listed above. The length of time a job requires to complete if it were to execute alone is called its execution time. The execution of a time-critical job cannot begin until its release time and must complete by its deadline. To validate a real-time system, its builder must demonstrate convincingly that all time-critical jobs will always complete by their deadlines, after making sure that the scheduler works correctly, that is, it never schedules any job before its release time. Hereafter, we assume 196 Sec. 9.1 Introduction 197 that the scheduler works correctly and focus on the problem of how to demonstrate that all jobs always complete by their deadlines. Traditionally, real-time systems have been validated by exhaustive simulation and testing. Because the execution time and release time of each job may vary, exhaustive methods are often infeasible or unreliable. This is especially true when the algorithm used to schedule the jobs is priority-driven. A scheduling algorithm is priority-driven if it does not leave any processor idle intentionally. Such an algorithm can be implemented by assigning priorities to jobs and placing all jobs that have been released and, therefore, ready for execution in a queue ordered by their priorities. The available …