Automating the deployment of 5G Network Slices with ONAP

TL;DR

This paper presents a model for automating 5G network slice deployment using ONAP, enabling faster, customizable, and private 5G network implementations.

Contribution

It proposes an ONAP-compatible network slice model based on 5G standards, demonstrating deployment of a private mobile core network.

Findings

Successful deployment of a private 5G network slice

Enhanced on-demand network provisioning capabilities

Accelerated time-to-market for 5G services

Abstract

Open Network Automation Platform (ONAP) is a carrier grade platform for automatically deploying and managing Virtualized Network Functions. In this paper, we address the deployment of network slices in order to come up with a model that is compatible with ONAP. We analyze various types of network slice ontology presented in the framework of 5G standardization bodies and we propose an ONAP-compatible model on the basis of which we illustrate the design, onboarding, instantiation and distribution of a network slice. We concretely define and deploy a network slice implementing a private and customized mobile core network. The achieved results not only make true NFV and 5G promises, notably those referring to on-demand networks, service customization and time-to-market acceleration, but they open the door to the deployment of private tailored cloud-native 5G networks.

Full text

Keywords: Network slicing, 5G, ONAP, NFV, SDN, On-demand Networks, Automation.

I Introduction

The concept of network slicing has become central in the design and deployment of 5G networks since it enables network operators to meet the specific service requirements expressed by vertical markets, such as real-time objectives, reliability and guaranteed Service Level Agreements (SLAs). 3GPP has notably introduced three macroscopic quality-oriented network slices namely enhanced Mobile BroadBand (eMBB), massive Machine Type Communications (mMTC), and Ultra-Reliable Low-Latency Communications (uRLLC). Beyond these three macro-service classes, network slicing can be seen as a means of dedicating network resources to specific purposes and/or customers. This approach takes benefit from new possibilities offered by virtualization technologies.

Network slicing based on virtualization can in fact be dated back to GENI initiative  [1] and notably the concepts presented by Feamster et al [2]. While network slicing was initially introduced to overcome the ossification of the Internet (e.g., via the implementation of new network protocols inside network slices), the emergence of Network Function Virtualization (NFV) [3] and Software Defined Network (SDN) as well as the massive deployment of Cloud infrastructures have radically changed the scope of network slicing.

As a matter of fact, one of the unprecedented innovations brought by virtualization technologies in the design of networks is the separation between network functions and their hosting hardware. Virtual Network Functions Virtualized Network Functions (VNFs) can thus be instantiated, on the fly, on common hardware and more generally on Cloud platforms. A network slice can then be defined as an independent virtual network running on top of a shared and/or dedicated infrastructure fulfilling specific performance requirements. A network slice is in fact a polymorphic entity which can be as simple as a Virtual Private Network (VPN) with a given SLA in terms of bandwidth and latency or a much more complex object embedding VNFs with more or less stringent isolation and performance requirements.

Network operators can actually take advantage of NFV and network slicing for offering portions of their networks to meet the needs of vertical industries as health-care, automotive, financial, etc. Network slicing is thus the capability of building packages of VNFs or more specifically Service Functions Chains Service Function Chains (SFCs) dedicated to specific purposes. For instance, a virtualized end-to-end mobile network including both core and radio access networks (with private antennas) can be composed and dedicated to a given company for setting up a private mobile network. Such a network slice can appear as an enriched VPN with a radio component. The major innovation of network slicing is then the flexibility, adaptability and time-to-market acceleration when deploying customized networks.

While both NFV and SDN respectively enable running network functions as applications and decoupling control and data functions, none of them manages the entire life-cycle of VNFs. To address this need, Open Network Automation Platform (ONAP) initiative was launched in 2017 promising automation of network services deployment as chains of VNFs. ONAP offers a number of features involved in the service design and life-cycle management, all working together [4]. For network operators, ONAP concretely enables the orchestration, control and operation of end-to-end network services as network slices. ONAP also promises advanced monitoring and analytic techniques to improve legacy network services.

In this paper, we address the problem of automating the deployment of network slices via ONAP. We present a unified network slicing model that gathers both ONAP software entities and 3GPP network architectures. On the basis of the proposed model, we present how to design and run a network slice that implements a private mobile core network to be dedicated to a given company. In addition, we identify some limitations in the current ONAP distribution and propose solutions to overcome these shortcomings.

We concretely present the design, on-boarding, instantiation and distribution of a fully virtualized 5G network slice which involves private core network functions and a shared access network. The exposed network slice is based on a complete separation of the data and control planes as recommended by 3GPP and known as Control User Plane Separation (CUPS). The deployed use-case makes true the 5G promises notably those referring to on-demand services and reduction of commercialization cycles. Results show that network slices can be instantiated on the fly and answer emerging markets needs.

This paper is organized as follows: In Section II, we review some definitions of network slicing given by the Industry and standardization bodies. In Section III, we present the proposed ontology approach and show that ontology provides models that are very relevant in the context of network slicing. In Section IV, we present a unified model compatible with ONAP. This model is illustrated in Section V in the case of a mobile core network deployed on demand. Some concluding remarks are presented in Section VI.

II Network Slicing: background and definitions

Network slicing is being widely studied in Academia and Industry since it is a promising concept and technique for customizing and dedicating services according to client needs based on virtualization. Customizing services and isolating networks is actually not new. We can for instance refer to label-based protocols that encapsulates traffic into virtual links, such as Multiprotocol Label Switching (MPLS) or IP tunnel-based paths performed by VPNs among various technologies allowing differentiated quality of service management.

The major innovation of network slicing comes together with NFV technologies. Both slicing and NFV enable customizing and deploying new network services on the fly while guarantee elasticity, scalability, and cost-efficiency. In other words, connectivity and specific treatments of traffic (both signaling and data) can be tailored to specific needs on demand. This claim enables operators to offer new services into an evolved framework of business models.

A network slice has been defined in the literature in many different ways: a bundle of services, a logical network, a type of virtual networking architecture, a chain of network functions created on top of a cloud infrastructure, etc.

According to the GSM Association [5], from a mobile operator point of view, a network slice is an independent end-to-end logical network on the top of a shared physical infrastructure with a negotiated QoS.

For Industry (see [6] for instance), network slicing creates separated use-case-specific logical networks upon a shared physical infrastructure. This is enabled by technology advances such as NFV, SDN and Management and Orchestration (MANO). Network slices share the same physical network infrastructure but are effectively distinct and isolated up to some extent.

China Mobile, Huawei, Deutsche Telekom, and Volkswagen have released their shared vision for the 5G era [7]. In their point of view, network slices are sets of dedicated logical Network-as-a-Service Network as a Service (NaaS) meeting requirements coming from vertical industries. Through flexible and customized design of functions, isolation mechanisms, and O&M tools, network slicing is capable of providing logical dedicated networks upon a common infrastructure.

For China Telecom [8], a network slice is an end-to-end logical subnet requiring the coordination of a core network, an access, and a transport network. Network slices can be based on isolated resources and/or shared resources. A network slice is deployed in a service-oriented architecture enabling network features configurations (in terms of QoS, system capacity, data rate).

Other definitions can be found in the technical literature, see for instance those presented by European projects [9, 10]. It turns out that a network slice is commonly viewed as a logical network embedding network functions and IT resources in order to meet specific business requirements by a customer.

In this context, a network slice is properly speaking not a service in itself; a slice is invoked to realize one or more services [6]. A network slice is then a logical network serving a defined business purpose with specific characteristics, and comprises all the required network resources. Such resources can be physical or virtual, and either dedicated to a particular slice, or shared between several slices [6]. The ETSI GR NGP in its Network Slicing Reference Framework [11] defines a network slice as a description of a service aware logical network that is composed of different physical or virtual network elements, resources and functions.

In the following section, to capture the complexity of network slicing, we use the concept of ontology. This subsequently allows us to come up with a model compatible with ONAP. The benefit of using an ontology approach is in that several models can be elaborated (possibly one model per operator) and are eventually composable in a global framework, which gives an enriched knowledge basis for formal reasoning.

III Slice ontology

Various network slicing ontologies are available in the literature, see for instance that proposed in [12]. Another interesting ontology is given by the ONAP community in the framework of the 5G slicing use case [13]. The model is based on 3GPP specifications and more concretely on the 3GPP TS.530 [14] (see Figure 1 for an illustration).

This ontology notably complies with the 5G architecture defined in 3GPP TS 23.501 [15]. It considers that communication service instances are provided by various Network Slice Instances (NSIs). In addition, it introduces the concept of network slice subnet, which is used by both 3GPP and ETSI. A complete slice is then the interconnection of slice subnets referred to as Network Slice Subnet Instances (NSSIs). A slice subnet is in turn composed of network functions. According to ETSI a subnet represents single or multiple networks under the control of an agent [11]. An agent makes reference to a logical entity that has complete control of its network infrastructure.

Figure 2 shows that different slices ($A$, $B$ and $C$) contain network functions belonging to the NSSI Access and Core Network. Communication Services are then provided through different NSIs [14]. The above ontology is purely descriptive and does not consider the hosting infrastructure, which shall enable the deployment of VNFs belonging to slices. In fact, the performance of network slices directly depends of VNF placement that compose them. For instance, slices requiring low latency performance need to be placed as close as possible to end users, network functions executing computation intensive tasks require high performance processors, etc. Thus, it is possible to develop another ontology taking into account all the processes involved in the creation of a slice. The goal is to develop a method of reasoning when dealing with multi-domain cases, i.e. when several actors (or operators) are involved in the creation of a slice.

For this purpose, we can consider the creation of a slice as a deployment exercise [16].

This approach aims at guaranteeing that the capabilities provided by systems can fulfill the needed capabilities.

The Open Net-centric Interoperability Standards for Training and Testing (ONISTT) approach is presented in [12]. The authors use Web Ontology Language (OWL) to express different types of ontology.

The strength of ONISTT is that it consider: (i) an analyzer that applies general logical reasoning and domain-specific rules to determine whether a candidate confederation can satisfy the requirements of a proposed event; (ii) the knowledge is captured from instance data into a Knowledge Base (KB). It turns out that ONISTT ontology concepts can be adapted to the context of network slicing. A top-level ONISTT-based ontology is shown in Figure 3.

To adapt the ONISTT approach to slicing, we use the generic term Network to design the classical network. That is a collection of network elements such as switches, routers, optical cross-connects, etc.

interconnected by transmission links and possibly controlled by SDN controllers) augmented by IT resources (disk, CPU, RAM) hosted in data centers. The slice ontology has three complementary parts that compose a Deployment (i.e., when instantiating a network slice):

• $\cdot$

A Bundle has Service objectives (e.g., an interconnection between origin-destination pairs such a classical VPN, a virtual EPC, etc.), from which an assemblage of capabilities is needed to offer all individual services composing the slice bundle.

• $\cdot$

A Network is a set of Resources offering individual capabilities, which can be combined to support the capabilities needed by a Service .

• $\cdot$

A set of Assignments match the capabilities offered by individual network resources with those required to support the service bundle.

The resulting ontology is illustrated in Figure 3 and covers all the aspects of a slice deployment. In the following section, by keeping in mind all the aspects of slice deployment, we focus on an ontology which is compatible with ONAP.

IV Unified Network Slicing Model

In this section, we continue along the line of defining a network slicing ontology by keeping in mind that it has to be compatible with the automation platform that shall orchestrate the network slices (namely, ONAP). The proposed model considers the required entities, actors and properties for automating the deployment of network slices. It is shown in Figure 4. Our proposal not only considers the slicing principles introduced by both 3GPP [14] and ETSI [11] but moreover the deployment requirements when considering the automation of the whole lifecycle of network slices, and more specifically when using ONAP.

We begin considering that a slice is by definition composed of a bundle of services offered by a single or several virtualized network functions running on shared or dedicated infrastructures. In general, a slice, beyond its polymorphic nature, is composed of a number of services to meet the global slice objectives/requirements. These requirements can be specified in a slice SLA detailing the objectives for the various services composing the slice (referred to as Network Slice Service Profile); each service can even have its own SLA and the SLAs of the various services composing a slice can be combined to achieve the SLA of the slice.

The proposed model involves the following entities:

• $\cdot$

Customer: Defines the customer properties (e.g., name, description, etc.). A customer is the entity that requires the deployment of a network slice and shall use it. A customer can be an enterprise, a single user, or a group of users belonging to a category (e.g., e-Health,).

• $\cdot$

Network Slice Provider (NSP): Deploys a network slice on the basis of network resources or services from multi-domain and/or multi-technology networks [11].

• $\cdot$

Vendor Software Product: This is the owner of an ONAP-resource. The defined ONAP resources are: Virtual Function (VF), Virtual Function Component (VFC), Complex Resource (CR), Physical Network Function (PNF), Connection Point (CP) Virtual Link (VL).

• $\cdot$

Network Service: Bundle of network functions that accomplishes a customer need. The granularity of a network service is given by the designer. For instance an end-to-end 5G network can be seen as a single service; however, both RAN and Core networks can be also viewed as individual services, namely RANaaS, 5GCNaaS. The final goal is to achieve loosely coupled services in order to improve modularity and individual deployment. Smaller services enable flexible, efficient and agile developments and deployments. The management of services become also more resilient.

• $\cdot$

Network Slice: Logical entity that is composed by a single or multiple network services (also referred to as subnets). When using the concept of slice subnet, a Network Slice Subnet (NSS) carries out a single service. In the following, we avoid using the subnet concept because it introduces unnecessary complexity and ambiguity. The granularity of a given service and its components shall generally be defined by the software product vendor who shall accomplish given guidelines (e.g., by network operators).

• $\cdot$

Network Function: Entity that accomplishes a task for a given network service. A network function can be a PNF or a VNF. A network function can in turn be composed of network sub-functions. The design of network functions and their components (referred to as VNFC) shall be determined by the software product provider.

• $\cdot$

Network Slice Service Profile: Contains the description of a customized service. From a business point of view, service profiles constitute the available slice offers that are or can be commercialized. A Network Slice Service Profile (NSSP) can then be offered to a potential slicing consumer as a basis on which clients can customize their slices. Certainly, offering standard NSSPs shall be enough to satisfy the majority of customers. Examples of network slice requirements that can be part of the NSSP are: area traffic capacity, charging, coverage area, degree of isolation, end-to-end latency, mobility, overall user density, priority, service availability, service reliability, UE speed, among others [14].

• $\cdot$

Network Slice Template: Technical description of a network slice. It contains attributes values for the creation and management of network slice. For a given network slice, the Slice Template specifies the charging capacity, guaranteed latency, guaranteed reliability, guaranteed data rate as well as the resource requirements (CPU, RAM, storage, network) for each network function composing the network service.

A slice Template is then directly related to HEAT files describing the network services that compose the underlying slice.

• $\cdot$

SLA: Contract established between a customer and the Network Slice provider according to the requirements defined in the NSSP and that shall be deployed under the principles of the Slice Template.

• $\cdot$

Physical link: Entity that represents physical connections between hosts, which can be located in distant geographic zones. According to ETSI [11] network links are a type of resource.

• $\cdot$

Host: Physical compute engine, which can support a single or multiple tenants.

• $\cdot$

Tenant: Isolated set of resources (compute, network, storage) belonging to a customer or to a network slice provider.

• $\cdot$

Virtual link: Enable the connection of VNFs via connection points.

• $\cdot$

Connection Points: Represents the network ports of VNFs. Connections points enable communicating two or more virtual network functions.

Virtual Links and Connection Points are rarely used. In fact, network ports, virtual networks and subnetworks are generally defined as part of network services or network functions. In ONAP, a network service is usually defined by a single HEAT template or composed of a forwarding graph of available network functions.

The various elements of the proposed slice ontology are illustrated in Figure 4. The grey boxes indicate those covered by ONAP. As we can observe many elements are missing in ONAP in order to have a holistic view of network slicing within the automation platform. We illustrate the above ontology in by a mobile core network.

V Network Slicing Automation: Illustration for a customized mobile core network

V-A Use case description

We consider the case of a company willing to deploy its own mobile core network on the top of a commercial mobile network, i.e., by using the radio access network of a network operator (say Grey Operator). The private core network (referred to as Slice A) can then implement customized policies in terms of security, reliability and quality of service (latency, data rate, etc).

The shared eNodeB is in charge of rerouting the private and public traffic to the adequate core network (namely, to the corresponding MME in 4G networks); see Figure 5 for an illustration. A routing solution has been proposed by the eDecor project [17]; we hence consider this issue as being solved. In the following, we focus on the network slice deployment, i.e., the onboarding and instantiation of a private mobile core network by using a virtualized infrastructure managed by ONAP.

The proposed network slice specifically implements a customized mobile core network based on various open-source solutions, notably Open Air Interface (OAI) code. The mobile core network considered in this paper is a fully virtualized 4G core network.

We specifically use the virtualized 4G core network implemented by b$<>$com’s solution [18]. It notably realizes a complete separation of the user and control planes as recommended by 3GPP for 5G networks, referred to as CUPS. The various components of the core network are connected via VxLANs (namely, MANAGEMENT, OVS-CTL, LTE-CTL,SECURE, CORE-CTL).

When a UE attaches to the private core network, the Authentication, Authorization, and Accounting (AAA) procedure is triggered by the Mobility Management Entity (MME). User profiles are validated by the Home Subscriber Server (HSS) data base which stores various parameters such as apn, imei, sim-card information OP, key, Mobile Network Code (MNC), Mobile Country Code (MCC), among others. When access is granted to the UE, the DHCP component provides it the IP-address, which is taken out of the address-pool established in the AAA component. The end-to-end connection is assured after the creation of the GTP-U and GTP-C tunnels. The NAT component provides address translation and is deployed between the SGi interface and the Internet network.

V-B Testbed settings

The onboarding and instantiation of the private mobile core network (Slice A) has been carried out with ONAP Casablanca 111ONAP is the result of the merge of two open source projects: Open Enhanced Control Orchestration Management and Policy (ECOMP) from AT&T and Open-O from Linux foundation MANO project. The ONAP project was formed in March, 2017 in response to a rising need for a common platform for telecommunication, cable, and cloud operators—and their solution providers—to deliver differentiated network services on demand, profitably and competitively, while leveraging existing investments [19].. We concretely create three private tenants on an Openstack-based platform.

The first tenant contains the ONAP subsystems and the two others are respectively reserved for the deployment of the control plane and user plane of the private core network. The testbed architecture is shown in Figure 6. The ONAP platform provides independent subsystems to design, create and manage the life-cycle of network services. Network slices can be seen as network services.

V-C Network Slice design

Generally speaking, ONAP considers a Network Service as a chain of VNF (also referred to as VF) which in turn can be formed by VNF Components (VNFCs). We define the Network Slice A (private mobile core network) as a chain of two independent services corresponding to the control and data plane functions. Each service is formed by a single VF, i.e., Core-CP-VF (control plane), Core-DP-VF (data plane). Each VF is in turn formed by various components. VFs are described by means of Heat Templates (one for each) and their components correspond to Openstack resources (OS::Nova::Server, OS::Neutron::Net,OS::Neutron::Subnet,OS::Neutron::Port).

The resulting service model to be deployed on ONAP is shown in Figure 7. The Grey Operator is then the Network Provider that deploys the Slice A demanded by a given company. Slice requirements are defined by the customer and included in the Network Slice Service Profile. The customer requirements are captured by the Network Slice Template which defines the technical aspects that enable deploying the demanded slice (e.g., network services, cloud owner, tenants, etc).

V-D Slice deployment workflow

In ONAP, the process for readying a service for distribution involves various roles based on a complex workflow. Before instantiating a network slice (an ONAP service) various stages need to be performed. They include onboarding, approving and deploying procedures which belong to the design-time environment of ONAP. The automation platform also supports run-time functions such as policy-driven automation and closed-loop management; these functions are out of the scope of the present paper.

For dealing with the various stages of the service onboarding and deployment, ONAP defines various roles: superuser, tester, designer, governor, and operator.

The workflow of a service deployment with ONAP is shown in Figure 8. The whole workflow is carried out by each service composing the network slice. Since the proposed model for the private mobile core network slice involves two services (Core-CP and Core-DP), the workflow has been twice performed.

The slice deployment begins with the validation of VFs. We assume that the images of the various VFCs exist. In practice, these images are created by the VNF provider. They can be in the form of .qcow2 in order to upload them onto an OpenStack platform. This latter task requires the preparation of Heat templates, which need to be ONAP-compatible (See Section V-E for details).

Thus, we build the ONAP-compatible HEAT templates for both Core Control-Plane and Core Data-Plane functions. Then, we create and certify them as ONAP VFs. For realizing the Core-CP-VF and Core-DP-VF for instantiation, we create, test and distribute the corresponding services. Finally, we instantiate the Core-CP and Core-DP services through the Virtual Infrastructure Deployment (VID) app of ONAP.

V-E Heat Template guidelines

ONAP defines stringent naming conventions and restricts the use of various Openstack resources such as Resource OS::Neutron::FloatingIP, Resource OS::Neutron::FloatingIPAssociation, among others. The ONAP requirements concerning the definition of HEAT files are detailed in [20]. In addition, ONAP requires mandatory metadata when defining nova-server resources. Metadata must include the following parameters: vnf_name, vnf_id, vf_module_id.

On the other hand, the ONAP environment file must not be greater than $2000$ characters (including quotes for each of them). In fact ONAP concatenates all environment variables to be inserted as a single database field (namely, Resource_input) into the vnf_resource_customization table during the service onboarding. This size will be upgraded to $20000$ in the Dublin version of ONAP, while for Casablanca a patch is required.

V-F Gap analysis

In ONAP, a network slice might be defined as one or more network services. However, an upper abstraction for managing these various network services that compose a network slice has not been defined so far. When a single network slice is defined by more than one network service (e.g., the proposed private core network which is composed of two services, Core-CP and Core-DP) the lifecycle management of each of them need to be performed independently.

The main drawback of defining a network slice as a network service is that the whole service shall be instantiated on a single hosting infrastructure (namely, an Openstack tenant). By definition in ONAP, a network service represents the finest granularity of placement. In other words, all VFs belonging to a network service shall be instantiated on the same tenant.

When considering the end-to-end performance (e.g., in terms of latency) of a given network service, some of their components (VFs) may require to be placed at different geographic locations. Ideally, VNFs should be located, where they are the most efficient in terms of performance and cost. The problem of VNF placement has been widely studied in the literature  [21, 22] in order to guarantee service requirements while optimizing cloud and network resources (i.e., bandwidth computing, memory, storage).

Even though the ONAP community has defined the ONAP Optimization Framework (OOF) module for dealing with placement, these functionality has not been fully exploited.

VI Conclusions

We have proposed a network slicing ontology compatible with ONAP. The proposed ontology takes into account major aspects of network slice deployment that we have illustrated by considering a network slice based on a private core network. This has enabled us to identify some shortcomings in the development of ONAP.

It turns out that in order to reach end-to-end performance requirements as well as greater flexibility and scalability, the components belonging to a network slice may require to be instantiated on different geographic locations (namely, various Openstack tenants or platforms) but seen by ONAP as a unique object. This in fact requires to introduce an intermediate abstraction layer for managing the various network services that compose a network slice. This abstraction layer should also enable selecting the hosting infrastructure of each network service component according to performance-based placement policies.

List of Acronyms

AAA Authentication, Authorization, and Accounting CP Connection Point CR Complex Resource CUPS Control User Plane Separation ECOMP Enhanced Control Orchestration Management and Policy eMBB enhanced Mobile BroadBand HSS Home Subscriber Server KB Knowledge Base MANO Management and Orchestration MCC Mobile Country Code MME Mobility Management Entity mMTC massive Machine Type Communications MNC Mobile Network Code MPLS Multiprotocol Label Switching NaaS Network as a Service NFV Network Function Virtualization NSI Network Slice Instance NSS Network Slice Subnet NSSI Network Slice Subnet Instance NSSP Network Slice Service Profile OAI Open Air Interface ONAP Open Network Automation Platform ONISTT Open Net-centric Interoperability Standards for Training and Testing OOF ONAP Optimization Framework OWL Web Ontology Language PNF Physical Network Function SDN Software Defined Network SFC Service Function Chain SLA Service Level Agreement uRLLC Ultra-Reliable Low-Latency Communications VF Virtual Function VFC Virtual Function Component VID Virtual Infrastructure Deployment VL Virtual Link VNF Virtualized Network Function VNFC VNF Component VPN Virtual Private Network