Bandwidth Provisioning for Network Slices with Per User QoS Guarantees

Abstract

A Network Slice (NS) is a set of network resources deployed to deliver premium service to a group of users as described in a Service Level Agreement (SLA). We consider that premium service entails per user Quality of Service (QoS) requirements instead of aggregate NS metrics as in the majority of the literature. Since the users in a NS may run applications with different QoS requirements, aggregate metrics are inadequate. Their inefficiency is even more pronounced in Radio Access Network Slices (RANSs). In this case, even if all users run the same application, provisioning bandwidth based on aggregate metrics may violate the QoS requirements of users experiencing poorer channel conditions. To resolve this issue, we propose a method to compute the bandwidth required to deliver per user average packet delay guarantees in enhanced Mobile Broadband (eMBB) RANSs by dimensioning a multiclass queueing system. Our method considers a varying number of NS users over time and allows the online adaptation of the NS bandwidth to its current traffic load, which significantly limits bandwidth overprovisioning. We also describe in detail how an SLA is reached between the tenant and the Network Operator (NO) based on the cdf of the required NS bandwidth. We conduct simulations to show the need for per user guarantees and the efficiency of our method.