GPGPU aided method for real-time systems

Guaranteeing real-time is required in the field of digital control. This is to start a task at a predetermined time, and complete the task before the predetermined deadline. However recently, it is getting difficult to guarantee such a completion. This is due to the fact that the amount of calculation within a period is increasing, and the period itself is being required to be shorter. As a method for parallel processing, GPGPU (General-Purpose computation on GPU) is recently attracting attention. However, tools to apply GPGPU to real-time system are not provided from the GPU vendor, and related research is not yet found. Therefore, this paper proposes and evaluates a method to operate GPGPU in real-time systems. From the experimental results, the proposed method dramatically reduced the jitter and the elapsed time comparing to the CPU itself, and also the conventional method. This shows that the proposed method improves the predictability of the completion time, and the GPU can be applied to real-time systems.