Optimal Computational Resource Allocation and Network Slicing Deployment in 5G Hybrid C-RAN

Abstract

Network virtualization is a key enabler for the 5G systems for supporting the novel use cases related to the vertical markets. In this context, we investigate the joint optimal deployment of Virtual Network Functions (VNFs) and the allocation of computational resources in a hybrid cloud infrastructure by taking into account the requirements of the 5G services and the characteristics of the cloud nodes. To achieve this goal, we analyze the relations between functional placement, computational requirements, and latency constraints, and formulate an integer linear programming problem, which can be solved by using a standard solver. Our results underline the advantages of a hybrid architecture over a standard solution with a central cloud, and show that the proposed mechanism to deploy VNFs leads to high resource utilization efficiency and large gains in terms of the number of slice chains that can be supported by the cloud-enhanced 5G networks.

Distributed Cloud Association and Beamforming in Downlink Multi-Cloud Radio Access Networks

“…We denote this cost as {P proc k , ∀k ∈ K}. More complicated computational cost models are considered in [33], where a normalized cubic function of the rate is assumed, a quadratic model is used in [34] and a linear model in [35]. Investigating various computational cost models for evaluating the overall performance is out of this work's scope and left for future investigations on this topic.…”

Optimal Virtual Network Function Deployment for 5G Network Slicing in a Hybrid Cloud Infrastructure

“…All the above works assumed slices with the same constraints; however, 5G systems need to comply with services with diverse requirements, which determine the computational and latency constraints of each VNF. In work [1], we investigated the deployment of VNF chains in a simple cloud infrastructure composed by one central cloud and one edge cloud.…”

“…, K} the set of edge clouds. Also, we denote the computational capacity [GFLOPS/s] of the k-th cloud as C k , the distance 1 of the k-th cloud from the r-th RRH as d k , and the distance between two clouds k, j ∈ K as 1 Although d r k could be a more precise notation, for sake of readability, we drop the index r. In fact, in this work we do not focus on the association between the VNF chains and the RRHs, and we assume that each chain is connected to a single RRH r ∈ R, which yields unique and well-defined d k . From a functional perspective, 5G BSs are decomposed into radio units (RUs), distributed units (DUs), and centralized units (CUs) [25], which can be implemented either as VNFs or the standard dedicated hardware (see the RAN slice subnet in Fig.…”

“…The model described in (1) It is well understood that the VNFs of the PHY layer, especially the encoding/decoding, are the most computational demanding functions in the cloud [5]. For a given VNF chain, the larger computational demand of lower layer VNFs with respect to functions related to upper layers is modelled by fitting coefficients in (1) with larger values [29].…”

“…The model described in (1) It is well understood that the VNFs of the PHY layer, especially the encoding/decoding, are the most computational demanding functions in the cloud [5]. For a given VNF chain, the larger computational demand of lower layer VNFs with respect to functions related to upper layers is modelled by fitting coefficients in (1) with larger values [29]. In addition, from (1), the computational requirement of a VNF increases with the spectral efficiency and the number of RBs, which together define the average throughput requirement of the slice chain; therefore, applications characterized by very high data rate, such as virtual/augmented reality, are more computational demanding than classical services such as eMBB or mMTC.…”