RT-Frontier: a real-time operating system for practical imprecise computation

Imprecise computation is known as an effective technique for dynamically resolving trade-offs between the amount of resources and the quality of the result. However, its implementation and operating system support methods have not been exploited enough from a practical point of view. This paper presents a new approach taken in the RT-Frontier operating system to support imprecise computation. Applications that allow imprecise computation are first transformed to tasks composed of three parts based on an extended imprecise computation model. All tasks are then uniformly scheduled according to a novel scheduling algorithm called Slack Stealer for Optional Parts (SS-OP). The SS-OP algorithm is designed to handle imprecise computations with small overhead, which is at a comparable level of that of the Earliest Deadline First (EOF) algorithm. The results of experiments show that the presented approach is cost-effective enough to be considered as a practical basis for embedded real-time systems.