Edge Function-as-a-Service is a new way of computing that dynamically schedules function executions across distributed edge (close to users) locations to reduce latency and improve user experience. For energy-efficient function scheduling that allows proactive resource allocation, accurate time-series prediction models that can predict how many times a function will be called in the future are very important. In this study, we assess the efficacy of various neural time-series predictors, specifically Gaussian processes, recurrent neural networks, and transformer architectures, in predicting the frequency of function invocations. We also suggest the Energy-Aware Resource Management (EA-RM) scheduling algorithm, which is based on a mixed-integer problem and aims to cut down on overall energy use by using fewer edge nodes. We look at how the accuracy of predictions affects function scheduling in terms of energy use, using real-world data that includes different functions and resources. The transformer-based predictor does better than the other predictors that were looked at in experiments, which leads to more accurate function scheduling. Also, the EA-RM algorithm for resource allocation cuts energy use by about 12–45% compared to competitors, and it has been shown to be more reliable. the precision of the employed prediction model.
Mrs.Ch.Swapna et al. (Wed,) studied this question.