Orchestrating real-time IoT workflows in a fog computing environment utilizing partial computations with end-to-end error propagation

With the explosive growth of the Internet of Things (IoT), fog computing emerged as a new paradigm, in an attempt to minimize network latency. Fog computing extends the cloud to the network edge, closer to where the IoT data are generated. Typically, fog resources are of limited capacity. On the other hand, IoT applications are becoming more and more complex and computationally demanding, requiring a certain level of Quality of Service (QoS) within strict time constraints. In such a real-time setting, it is often more desirable for a job to meet its deadline by producing an approximate—but still of acceptable quality—result, rather than producing an overdue precise result. Based on this concept, in this paper we examine the orchestration of real-time IoT workflows in a heterogeneous fog computing environment, utilizing partial computations. When a workflow task produces an imprecise result, the error may be propagated not only to its immediate child tasks, but also across subsequent successor tasks of the workflow, ultimately affecting its end-result. The proposed scheduling technique is compared to a baseline algorithm, where partial computations are not used, under various result precision thresholds and input error propagation probabilities. The simulation results reveal that the proposed heuristic can provide on average a 32.71% lower deadline miss ratio than the baseline policy, by trading off an average result precision of 2.43%.