Evolution of access network sharing towards software based passive optical network: can virtual OLTs run at full line rate? Marco Ruffini (1) , David Coyle (2) , Jasvinder Singh (2) , Rory Sexton (2) , Brendan Ryan (2) , Robin Giller (2) , Michael O’Hanlon (2) , Frank Slyne (1) Abstract—This paper summarises how access network sharing has evolved from physical layer unbundling to fully virtualised architectures. We address one of the crucial question in access network virtualisation: whether fully virtualised PONs can run at line rate over general purpose processing units. This paper provides an experimental insight into this query, showing that it is possible to run XGS-PON at line rate, if additional hardware acceleration is provided, e.g., for cryptographic functions. Index Terms—DBA, virtualization, PON, NFV, SDN, access I. I NTRODUCTION Networks are fundamentally built on the notion of sharing. Starting from the concept of statistical multiplexing of dif- ferent users and applications, over time, through deregulation and increase in competition, the concept of sharing was also implemented across network providers, which would share infrastructure by operating multi-tenancy solutions. The past 25 years have seen, for example, access networks sharing evolving through multiple steps. The first sharing technology, the Local Loop Unbundling (LLU), focused on physical layer access, as Other Licensed Operators (OLOs) needed to get physical access to end user lines at each central office where service was provided. Although this removes the high cost part due to cabling, still required presence at each central office. Thus other virtual sharing options started to appear, in order to reduce cost for new entrants, so that OLOs could provide services to end users, by having their customers’ data aggregated at regional Points of Presence (PoPs). Different technologies were developed, namely Bit- srtream, Virtual Unbundling Local Access (VULA) and next generation bitstream, offering different trade-offs between cost and ability to control the infrastructure. For example, LLU allows the highest level of control as new entrants operate their own physical layer equipment, but it requires new entrants to provide equipment locally, at each CO they operate, which is expensive. Bitstream instead enable entrants to use most of the equipment from incumbent operators, but have no control on the service quality. VULA and NGA bitstream have tried to provide some level of mitigation, by providing additional QoS differentiation. The work in [1] provides a survey of the evolution of such sharing mechanisms. In the meantime, server virtualisition started to be applied in the computing domain, showing the potential that such 1 CONNECT Centre, Trinity College Dublin marco.ruffini at tcd.ie 2 Intel Corporation Ireland. technology could bring to infrastructure sharing. Thus, as the concept of function virtualisation made its way in the networking world, we has seen the development of a com- pletely new model for sharing access network resources. The Central Office Rearchitected as a Data Centre (CORD) [2], adopts the concept of network functions virtualisation and scales it up to implement all the main functions typically found in a central office. The code implementation of this open source project, coordinated by the Open Networking Foundation (ONF), focused on three complementary use cases: residential broadband, mostly delivered through Passive Op- tical Networks, Mobile Broadband, making use of Software Defined Radio (SDR) technology, and enterprise, focusing on metro transport. Such use cases are continuously evolving, and the reader can refer to [3] for the latest developments. II. DEEP PON VIRTUALISTION Concepts such as CORD were developed to provide full control to the OLOs, now better referred to as Virtual Network Operators (VNOs), for delivery of broadband services. Taking PONs as an example, virtualisation enabled dynamic sharing of resources across VNOs, which could take a share of the net- work performance and deliver arbitrary amounts of committed and peak data rate to its users. It should be noticed however, that in CORD the virtualisation is applied to the management layers of the system, i.e., VLAN configuration and setting of QoS parameters such as priority queues and Dynamic Band- width Allocation (DBA) assured and non-assured bandwidth values. The MAC layer however remains in hardware, meaning that this part is not virtualised and there is one common DBA mechanism that handles the scheduling for all VNOs. The main reason behind this choice was the computational cost of implementing the MAC in software (consider that this work originated in 2015). However, this mechanism is not applicable to future PON use cases that require low latency. Today for example, PONs are being increasingly considered for supporting Cloud-RAN and other low latency services, where controlling packet scheduling becomes essential to meet strict latency requirements. For example, this was experienced in the cooperative DBA [4] approach (standardised in [5]), which enables exchange of information between the C-RAN Distributed Unit (DU) and the OLT DBA mechanism, so that upstream scheduling can be coordinated in advance and the system can achieve the strict latency requirements needed for transporting the LLS interface. This, for example, cannot be 978-3-903176-21-8 © 2020 IFIP