Communication and Computing Cost Optimization of Meshed Hierarchical NFV Datacenters

Binayak Kar,Eric Hsiao-Kuang Wu,Ying-Dar Lin

doi:10.1109/access.2020.2995384

Abstract

Telecommunication carriers of 5G-MEC are re-architecting their central offices and mobile base stations as datacenters with network function virtualization (NFV) technology. These datacenters (DCs) are known as edge datacenters that help network operators speed deployment and reduce costs. Previously, the use of NFV was limited to within a datacenter (DC) known as intra-DC. Recently, many studies have been conducted into the use of NFV across DCs, i.e., inter-DC. However, these NFV inter-DC architectures have limited communication between DCs with either horizontal or vertical connectivity. In this paper, we propose a generic architecture of such edge NFV datacenters with both horizontal and vertical connectivity, and demonstrate the consequences of both vertical and horizontal connectivity between DCs in terms of communication and computing costs. We formulate a cost optimization problem with latency and capacity as constraints by estimating the traffic dispatch rate between DCs. We propose a vertical-horizontal communication (VHC) heuristic solution to the NP-hard problem. Compared to horizontal connectivity, our results show that vertical connectivity helps to reduce computing costs by 10-30%. However, both vertical and horizontal communications together can help to reduce such costs by 20-30% compared to only vertical communication.

Highlights

Network function virtualization (NFV) [1] is a new alternative technology in the revolution of the communication network that has emerged as an appealing solution to transform dedicated hardware implementations to software instances running in a virtualized environment
We considered five different types of topologies: tree (T), fat-tree (FT), partial mesh (M), tree with partial mesh (TwM), and fat-tree with partial mesh (FTwM)
When multiple flows shared a Virtual Network Functions (VNF), the cost was estimated based on when the first packet of the first arrival flow to that VNF started processing and when the last packet was served by that VNF

Summary

Introduction

Network function virtualization (NFV) [1] is a new alternative technology in the revolution of the communication network that has emerged as an appealing solution to transform dedicated hardware implementations to software instances running in a virtualized environment. In NFV, a requested service is implemented by a sequence of Virtual Network Functions (VNF) that can run on generic servers by leveraging virtualization technology These VNFs are pitched with a predefined order. In hierarchical connectivity, a parent DC may have multiple child DCs and a child DC can have multiple parents Taking these factors into account, we have considered five different topologies such as the partial mesh (M), tree (T), fat-tree (FT) [22], tree with partial mesh (TwM), and fat-tree with partial mesh (FTwM) that demonstrate the degree of DC connectivity (There are two kinds of degree of connectivity: (1) The horizontal degree of connectivity defines a node horizontally and directly connected to how many of its siblings (2) The vertical degree of connectivity means a node vertically and directly connected to how many of its parent nodes.) in all possible scenarios.

Objectives

Results

Conclusion