Edge Computing Sever Selection in Fog Radio Access Networks

Abstract

In this paper, we consider a fog radio access (F-RAN) network formed by two different kinds of node, namely the user and calculator (i.e., the edge computing sever), respectively. It is assumed that the users and calculators follows two independent homogeneous Poisson point processes (HPPPs). Further, it is assumed that the arrival of the computing tasks at each user and the departure of the computing tasks at each calculator follow two Poisson processes with rate ηu and ηc, respectively. That is, the time between successive arrivals at each user or departures at each calculator are independent exponential random variables with mean given by $\frac{1}{{{\eta _u}}}$ or $\frac{1}{{{\eta _c}}}$. Time is divided into slots. At the beginning each time slot, each user is designed to offload the computing tasks generated in the last time slot to their nearby calculators within a distance of Rd for data processing. Different from the random-chosen based calculator selection schemes studied in the previous works, in this paper, we consider a prediction based calculator selection strategy to enhance the delay performance of the F-RAN. Particularly, under the proposed prediction based calculator selection strategy, we estimate the network status in the next time slot, and use which as the criterion for computing server selection and tasks offloading. It is shown through both the analysis and simulations that the proposed prediction based calculator selection strategy outperforms the random-chosen based calculator selection protocol in terms of the successful offloading probability.