Experience with Implementation of a Turbojet Engine Control Program on a Multiprocessor

A control program for a turbojet engine has been implemented on a four processor computer, achieving a speedup of 3.38 times the speed of a sequential version of the same program on a single processor. The concurrent program was produced from a sequential program by subjecting the sequential program to global, hierarchical interprocedural data-flow analysis and timing measurements. A static schedule for the constituent tasks of the control program on the four processors was determined using a heuristic algorithm based on the critical path method. The program is implemented on a bus-based multiprocessor with private memory for each processor´s code and data, and shared memory for data passed between processors. An application-independent supervisor program on each processor controls the execution of tasks by ensuring that all data dependencies are met and copying values from shared memory before execution of each task, then copying values to shared memory and notifying tasks on other processors after a task has completed execution. This work is significant for the largely automated transformation of a sequential program via coarse-grain data-flow analysis to an efficient parallel implementation. The approach should be applicable to a variety of control and related programs where iterative tasks with well-bounded execution times are computed in systems with hard real-time requirements.