A Learning-based Framework For Edge Computing Resources Planning

With the development of the Internet of things and the increasing number of intelligent terminal devices,the amount of data in the network and the computing requirements of devices grow explosively,which puts forward higher requirements for the network performance,such as the data transmission delay,the number of terminal device connections,the energy consumption of the system,and the data security of users.However,traditional network is far from being able to meet these requirements.On the one hand,the network can not support so many device connections and massive data will cause serious congestion,which will further deteriorate delay performance.As a new computing mode,edge cloud computing can bring resources such as storage and computing closer to the edge of the network.Part of the user requirements can be processed at the edge of the network without being transferred to the core network,which not only reduces the backhaul network pressure,but also reduces the users' delay and supports delay sensitive applications.Therefore,the deployment of edge cloud platform can be regarded as the main solution to solve the future network transmission and computing pressure.However,as the first step to realize edge cloud,the deployment of edge cloud platform has not been deeply studied.Previous research on edge cloud deployment is mainly based on fixed resource requirements,ignoring numerous uncertain factors imposed by the real dynamic world.In view of this,we focus on the resource planning of edge cloud platform under different scenarios in uncertain environment.The main research contents and innovative work are summarized as follows:1)In mobile networks,the resource requirements of base stations show arbitrary changes due to the complexity of the real dynamic world,which brings challenges to the deployment of edge cloud platform.In order to optimize the resource optimization of mobile edge computing platform,we propose a server deployment model based on uncertain programming theory,which is modeled as a joint optimization problem of MEC edge server placement and resource allocation with uncertain base station computing requirements,with the goal is to minimize the data delay in the network.Owing to the complexity of the joint uncertainty problem,we design a learning-based algorithm framework,which combines genetic algorithm,Monte Carlo stochastic simulation and neural network,and simplex method is introduced to solve the sub optimization problem of server resource allocation.Finally,small-scale and large-scale simulations are implemented based on lognormal distribution base station traffic model and real data set respectively to evaluate the performance of our proposed framework.The results show that the proposed method can reduce the overall delay of the system under arbitrary base station traffic mode.2)In the Internet of things scenario,we propose the corresponding fog nodes deployment scheme in wireless sensor network.Due to the change of environment state,the data uploaded by sensor nodes to fog nodes also changes dynamically,which brings difficulties to the deployment of fog nodes.Similar to MEC server deployment,we also propose a fog node deployment model based on uncertain programming theory with the aim to reduce the overall energy consumption of the system by minimizing the transmission power of the system.We also adopt the learning-based algorithm framework to solve this uncertain programming problem,in which the sub optimization problem of fog node bandwidth assignment is a knapsack problem that can be solved by dynamic programming.In order to evaluate the performance of this scheme,we assess the fog node deployment framework based on the ON/OFF traffic model.The results show that this scheme can significantly reduce the system energy consumption compared with other algorithms under the same network conditions.