LRA-3C: Learning Based Resource Allocation for Communication-Computing-Caching Systems

Abstract

The diversified usage scenarios of 5G networks require to allocate communication-computing-caching (3C) multidimensional resources efficiently according to different service requirement in terms of latency, bandwidth, and connectivity. In this paper, a network slicing (NS) architecture based mobile edge computing (MEC) and software-defined network (SDN) technologies are proposed to support flexible 3C resource allocation for improving the service of three 5G typical usage scenarios, namely enhanced mobile broadband (eMBB), ultra-reliable and low latency communications (URLLC) and massive machine-type communications (mMTC). Furthermore, a neural network (NN) based 3C resource allocation algorithm is designed to provide resource allocation decision for the NS architecture. With the aid of data pre-processing techniques, fast resource allocation decisions can be made by the NN aided NS architecture. Meanwhile, the performance of the proposed NS architecture is investigated in a testbed environment. Experimental results obtained from the testbed demonstrate that the system performance of the NN aided NS architecture can be improved by function fitting based pre-processing approaches. Moreover, an accurate classification performance can be obtained by the proposed architecture for facilitating 3C resource allocations.