Service availability in the NFV Virtualised Evolved Packet Core Andres Gonzalez, P˚ al Grønsund, Kashif Mahmood Telenor Research Telenor ASA Email: {andres.gonzalez, pal.gronsund, kashif.mahmood}@telenor.com Bjarne Helvik, Poul Heegaard, Gianfranco Nencioni Department of Telematics Norwegian University of Science and Technology Email: {bjarne, poul.heegaard, gianfranco.nencioni}@item.ntnu.no Abstract—Network Function Virtualization (NFV) promises to transform the way telecom providers design and operate networks and network services. Virtualized Evolved Packet Core vEPC is one of the Network Function Virtualization NFV use cases that has got most of attention, where dependability is a major concern. In the traditional EPC, functions are deployed in proprietary network elements with proven characteristics, e.g., a defined availability, and corresponding guarantees. Hence, network operators have a firm basis for the design of a robust mobile core network. On the other hand, in the vEPC, network operators face a more challenging environment, where functions, subsystems and requirements are interrelated in a more complex manner. Hence, the assessment of the robustness of the network, and the design to meet dependability requirements becomes hard. In order to address this challenge, we provide initial guidelines and modeling tools to assess system availability in vEPC scenarios, and identify the most relevant factors to be considered in this process. Index Terms—NFV; virtual EPC; fault tolerance; mobile core reliability;availability modeling. I. INTRODUCTION The mobile networks have evolved significantly, moving from 1G to the current 4G networks. Common to all genera- tions is the proprietary, mission specific and topologically fixed nodes that provide the network functions. This has resulted, amongst other issues, in high CAPEX and OPEX for the operators and long lead times for changes in the service delivery. One solution to address this challenge is to virtualize the network functions and provide them on Commercial-Off- The-Shelf (COTS) hardware, i.e. to introduce network function visualization (NFV), where the aim is to run mobile network functions as software instances on commodity servers. An immediate advantage of such an approach is the ability to scale the resources up and down according to the traffic demand. This approach is highly relevant for network functions that do not require specific hardware (ASICs). The evolved packet core (EPC) in 4G networks consists of many such network functions and the potential benefits of a virtualized EPC (vEPC) are huge [2]. One of the main concern for the Telecom operators, and the focus of this paper, is the system availability in the vEPC. A firm control of this aspect is mandatory in order to consider an production scale implementation. To establish availability guarantees, responsibility domains and modeling tools in a vEPC environment is much more complex than in the traditional network design. The main questions addressed in this paper are: How to assess the availability of a vEPC? and What are the main availability concerns to be considered? For this, we study the potential failure sources in a vEPC environment, discuss the advantages and disadvantages of the vEPC from the dependability point of view, and finally, we propose a model based in stochastic activity networks, to assess the availability of a vEPC. There has been a considerable effort in dealing reliability and availability aspects in NFV, a significant part of it within the ETSI Industry Specification Group NFV REL WG [5]. They have established a set of requirements and specifications for developing robust NFV based services and have defined a number of techniques and mechanisms to ensure reliability and availability in an operational virtual environment. The work presented in this paper complements the defined specifications and requirements, by providing a Stochastic Activity Network (SAN) system availability model, and a study of the main parameters to be considered in such assessment. Virtualization is a technology that started in the 1970’s [10]. It has acquired significant importance for computational applications with the development of cloud computing, and now, it is extending its scope to the network domain. One of the arguments motivating the implementation of virtualized environments is the independence from the physical hardware, since any affected virtual application may be executed on any working server [15]. Fault tolerance is a key aspect in a virtualized environment wherein two most relevant techniques are active-replication and standby-replication [3], [4]. The availability of the datacenter network is fundamental for the vEPC robustness. There are several studies that have dealt with this. For e.g [9] presents a large-scale analysis of network failures [9]. They distinguish between two types of failures, link and device failures and observe that low-cost, commodity switches (such as ToRs and AggS) are highly reliable, but middleboxes (such as load balancers) experience a high number of software faults. Datacenter networks can be highly reliable, although issues as network control, redundancy and topology design should be carefully planned. Internal network failures in data centers, and network failures between data centers, have a significant impact on the virtualized