Scheduling of accuracy-constrained real-time systems in dynamic environments

Many real-time embedded systems are sensitive to both the accuracy and timeliness of job results. In this letter, two sources of inaccuracy are considered for such systems: 1) input data noise (IDN) due to the environmental transient noises, and 2) age of data (AD) related to the time of capturing data, which may depend on the length of time between capturing and using the input data. Thus, in the presence of one or more jobs in the system, some tradeoffs are needed among capturing data with an appropriate IDN when the environment is less noisy, reducing AD, and respecting the timeliness of jobs. Our emphasis in the current study is to model firm real-time jobs having some thresholds for the inaccuracy and handle the aforementioned tradeoff by the system scheduler. An online accuracy-aware real-time scheduler is also proposed and evaluated.