As Network Functions Virtualization (NFV) technology matures, multiple NFV orchestration solutions have emerged and 2016 was a busy year. While some commercial products were already available on the market, multiple open source initiatives were also launched, with most delivering initial code releases, and others planning to roll-out software artifacts later this year. With so much going on, we thought we'd provide you with a technical overview of some of the various NFV orchestration options, so you can get a feel for what's right for you. In particular, we'll cover:
In addition, multiple NFV projects have been funded under European Union R&D programs. Projects such as OpenBaton, T-NOVA/TeNor and SONATA have their codebases available in public repos, but industry support, involvement of external contributors and further sustainability might be a challenging for these projects, so for now we'll consider them out of scope for this post, where we'll review and compare orchestration projects across the following areas:
General overview and current project state
Compliance with NFV MANO reference architecture
NSD definition approach
VIM and VNFM support
Capabilities to provision End to End service
Interaction with relevant standardization bodies and communities
General overview and current project state
We’ll start with a general overview of each project, along with, its ambitions, development approach, the involved community, and related information.
The OpenSource MANO project was officially launched at the World Mobile Congress (WMC) in 2016. Starting with several founding members, including Mirantis, Telefónica, BT, Canonical, Intel, RIFT.io, Telekom Austria Group and Telenor, the OSM community now includes 55 different organisations. The OSM project is hosted at ETSI facilities and targets delivering an open source management and orchestration (MANO) stack closely aligned with the ETSI NFV reference architecture. OSM issued two releases,Rel 0 and Rel 1, during 2016. The most recent at the time of this writing, OSM Rel. 1, has been publicly available since October, 2016, and can be downloaded from the official website. Project governance is managed via several groups, including the Technical Steering group responsible for OSM's technical aspects, the Leadership group, and the End User Advisory group. You can find more details about OSM project may be found at the official Wiki.
The OPEN-O project is hosted by the Linux foundation and was also formally announced at 2016 MWC. Initial project advocates were mostly from Asian companies, such as Huawei, ZTE and China Mobile. Eventually, the project project got further support from Brocade, Ericsson, GigaSpaces, Intel and others. The main project objective is to enable end-to-end service agility across multiple domains using unified platform for NFV and SDN orchestration. The OPEN-O project delivered its first release in November, 2016 plans to roll-out future releases in a 6 month cycle. Overall project governance is managed by the project Board, with technology-specific issues managed by the Technical Steering Committee. You can find more general details about the OPEN-O project may be found at the project web-site.
Originally CORD (Central Office Re-architected as a Datacenter) was introduced as one of the use cases for the ONOS SDN Controller, but it grew-up into a separate project under ON.Lab governance. (ON.Lab recently merged with the Open Networking Foundation.) The ultimate project goal is to combine NFV, SDN and the elasticity of commodity clouds to bring datacenter economics and cloud agility to the Telco Central Office. The reference implementation of CORD combines commodity servers, white-box switches, and disaggregated access technologies with open source software to provide an extensible service delivery platform. CORD Rel.1 and Rel.2 integrate a number of open source projects, such as ONOS to manage SDN infrastructure, OpenStack to deploy NFV workloads, and XOS as a service orchestrator. To reflect use cases' uniqueness, CORD introduces a number of service profiles, such as Mobile (M-CORD), Residential (R-CORD), and Enterprise (E-CORD). You can find more details about CORD project can be found at the official project web site.
Gigaspaces’ Cloudify is the open source TOSCA-based cloud orchestration software platform. Originally introduced as a pure cloud orchestration solution (similar to OpenStack HEAT), the platform was further expanded to include NFV-related use cases, and the Cloudify Telecom Edition emerged. Considering its original platform purpose, Cloudify has an extensible architecture and can interact with multiple IaaS/PaaS providers such as AWS, OpenStack, Microsoft Azure and so on. Overall, Cloudify software is open source under the Apache 2 license and the source code is hosted in a public repository. While the Cloudify platform is open source and welcomes community contributions, the overall project roadmap is defined by Gigaspaces. You can find more details about the Cloudify platform at the official web site.
Compliance with ETSI NFV MANO reference architecture
At the time of this writing, a number of alternatives and specific approaches, such as Lifecycle Service Orchestration (LSO) from Metro Ethernet Forum, have emerged, huge industry support and involvement has helped to promote ETSI NFV Management and Orchestration (MANO) as the de-facto reference NFV architecture. From this standpoint, NFV MANO provides comprehensive guidance for entities, reference points and workflows to be implemented by appropriate NFV platforms (fig. 1):Figure 1 - ETSI NFV MANO reference architecture
As this project is hosted by ETSI, the OSM community tries to be compliant with the ETSI NFV MANO reference architecture, respecting appropriate IFA working group specifications. Key reference points, such as Or-Vnfm and Or-Vi might be identified within OSM components. The VNF and Network Service (NS) catalog are explicitly present in an OSM service orchestrator (SO) component. Meanwhile, a lot of further development efforts are planned to reach feature parity with currently specified features and interfaces.
While the OPEN-O project in general has no objective to be compliant with NFV MANO, the NFVO component of OPEN-O is aligned with an ETSI reference model, and all key MANO elements, such as VNFM and VIM might be found in an NFVO architecture. Moreover, the scope of the OPEN-O project goes beyond just NFV orchestration, and as a result goes beyond the scope identified by the ETSI NFV MANO reference architecture. One important piece of this project relates to SDN-based networking services provisioning and orchestration, which might be further used either in conjunction with NFV services or as a standalone feature.
Since its invention, CORD has defined its own reference architecture and cross-component communication logic. The reference CORD implementation is very OpenFlow-centric around ONOS, the orchestration component (XOS), and whitebox hardware. Technically, most of the CORD building blocks might be mapped to MANO-defined NFVI, VIM and VNFM, but this is incidental; the overall architectural approach defined by ETSI MANO, as well as the appropriate reference points and interfaces were not considered in scope by the CORD community. Similar to OPEN-O, the scope of this project goes beyond just NFV services provisioning. Instead, NFV services provisioning is considered as one of the several possible use cases for the CORD platform.
The original focus of the Cloudify platform was orchestration of application deployment in a cloud. Later, when the NFV use case emerged, the Telecom Edition of the Cloudify platform was delivered. This platform combines both NFVO and generic VNFM components of the MANO defined entities (fig. 2).Figure 2 - Cloudify in relation to a NFV MANO reference architectureBy its very nature, Cloudify Blueprints might be considered as the NS and VNF catalog entities defined by MANO. Meanwhile, some interfaces and actions specified by the NFV IFA subgroup are not present or considered as out of scope for the Cloudify platform. From this standpoint, you could say that Cloudify is aligned with the MANO reference architecture but not fully compliant.
Software architecture and components
As you might expect, all NFV Orchestration solutions are complex integrated software platforms combined from multiple components.
The Open Source MANO (OSM) project consists of 3 basic components (fig. 3):Figure 3 - OSM project architecture
The Service Orchestrator (SO), responsible for end-to-end service orchestration and provisioning. The SO stores the VNF definitions and NS catalogs, manages workflow of the service deployment and can query the status of already deployed services. OSM integrates the rift.io orchestration engine as an SO.
The Resource Orchestrator (RO) is used to provision services over a particular IaaS provider in a given location. At the time if this writing, the RO component is capable of deploying networking services over OpenStack, VMware, and OpenVIM. The SO and RO components can be jointly mapped to the NFVO entity in the ETSI MANO architecture
The VNF Configuration and Abstraction (VCA) module performs the initial VNF configuration using Juju Charms. Considering this purpose, the VCA module can be considered as a generic VNFM with a limited feature set.
From a general standpoint, the complete OPEN-O software architecture can be split into 5 component groups (Fig.4):Figure 4 - OPEN-O project software architecture
Common service: Consists of shared services used by all other components.
Common TOSCA: Provides TOSCA-related features such as NSD catalog management, NSD definition parsing, workflow execution, and so on; this component is based on the ARIA TOSCA project.
Global Service Orchestrator (GSO): As the name suggests, this group provides overall lifecycle management of the end-to-end service.
SDN Orchestrator (SDN-O): Provides abstraction and lifecycle management of SDN services; an essential piece of this block are the SDN drivers, which provide device-specific modules for communication with a particular device or SDN controller.
NFV Orchestrator (NFV-O): This group provides NFV services instantiation and lifecycle management.
The OPEN-O project uses a microservices-based architecture, and consists of more than 20 microservices. The central platform element is the Microservice Bus, which is the core microservice of the Common Service components group. Each platform component should register with this bus. During registration, each microservice specifies exposed APIs and endpoint addresses. As a result, the overall software architecture is flexible and can be easily extended with additional modules. OPEN-O Rel. 1 consists of both Java and python-based microservices.
As mentioned above, CORD was introduced originally as an ONOS application, but grew into a standalone platform that covers both ONOS-managed SDN regions and service orchestration entities implemented by XOS. Both ONOS and XOS provide a service framework to enable the Everything-as-a-Service (XaaS) concept. Thus, the reference CORD implementation consists of both a hardware Pod (consisting of whitebox switches and servers) and a software platform (such as ONOS or XOS with appropriate applications). From the software standpoint, the CORD platform implements an agent or driver-based approach in which XOS ensures that each registered driver used for a particular service is in an operational state (Fig. 5):Figure 5 - CORD platform architectureThe CORD reference implementation consists of Java (ONOS and its applications) and python (XOS) software stacks. Additionally, Ansible is heavily used by the CORD for automation and configuration management
From the high-level perspective, platform consists of several different pieces, as you can see in figure 6:Figure 6 - Cloudify platform architecture
Cloudify Manager is the orchestrator that performs deployment and lifecycle management of the applications or NSDs described in the templates, called blueprints.
The Cloudify Agents are used to manage workflow execution via an appropriate plugin.
To provide overall lifecycle management, Cloudify integrates third-party components such as:
Elasticsearch, used as a data store of the deployment state, including runtime data and logs data coming from various platform components.
Logstash, used to process log information coming from platform components and agents.
Riemann, used as a policy engine to process runtime decisions about availability, SLA and overall monitoring.
RabbitMQ, used as an async transport for communication among all platform components, including remote agents.
The orchestration functionality itself is provided by the ARIA TOSCA project, which defines the TOSCA-based blueprint format and deployment workflow engine. Cloudify “native” components and plugins are python applications.
Approach for NSD definition
The Network Service Descriptor (NSD) specifies components and the relations between them to be deployed on the top of the IaaS during the NFV service instantiation. Orchestration platforms typically use some templating language to define NSDs. While the industry in general considers TOSCA as a de-facto standard to define NSDs, alternative approaches are also available across various platforms.
OSM follows the official MANO specification, which has definitions both for NSDs and VNF Descriptors (VNFD). To define NSD templates, YAML-based documents are used. NSD is processed by the OSM Service Orchestrator to instantiate a Network Service, which itself might include VNFs, Forwarding Graphs, and Links between them. A VNFD is a deployment template that specifies a VNF in terms of deployment and operational behaviour requirements. Additionally VNFD specifies connections between Virtual Deployment Units (VDUs) using the internal Virtual Links (VLs). Each VDU in an OSM presentation relates to a VM or a Container. OSM uses archived format both for NSD and VNFD. This archive consists of the service/VNF description, initial configuration scripts and other auxiliary details. You can find more information about OSM NSD/VNFD structure at the official website.
In OPEN-O, the TOSCA-based templates is used to describe the NS/VNF Package. Both the TOSCA general service profile and the more recent NFV profile can be used for NSD/VNFD, which is further packaged according to the the Cloud Service Archive (CSAR) format. A CSAR is a zip archive that contains at least two directories: TOSCA-Metadata and Definitions. The TOSCA-Metadata directory contains information that describes the content of the CSAR and is referred to as the TOSCA metafile. The Definitions directory contains one or more TOSCA Definitions documents. These Definitions documents contain definitions of the cloud application to be deployed during CSAR processing. More details about OPEN-O NSD/VNFD definitions may be found at the official web site.
To define a new CORD service, you need to define both TOSCA-based templates and Python-based software components. Particularly when adding a new service, depending on its nature, you might alter one of several platform elements:
TOSCA service definition files, appropriate models, specified as YAML text files
REST APIs models, specified in Python
XOS models, implemented as a django application
Synchronizers, used to ensure the Service instantiated correctly and transitioned to the required state.
To instantiate a service or application, Cloudify uses templates called “Blueprints” which are effectively orchestration and deployment plans. Blueprints are specified in the form of TOSCA YAML files and describe the service topology as a set of nodes, relationships, dependencies, instantiation and configuration settings, monitoring, and maintenance. Other than the YAML itself, a Blueprint can include multiple external resources such as configuration and installation scripts (or Puppet Manifests, or Chef Recipes, and so on) and basically any other resource required to run the application. You can find more details about the structure of Blueprints here.
VNFM and VIM support
NFV service deployment is performed on the appropriate IaaS, which itself is a set of virtualized compute, network and storage resources. The ETSI MANO reference architecture identifies a component to manage these virtualized resources. This component is referred to as the Virtual Infrastructure Manager (VIM). Traditionally, the open source community treats OpenStack/KVM as a “de-facto” standard VIM. However, an NFV service might be span across various VIM types and various hypervisors. Thus multi-VIM support is a common requirement for an Orchestration engine. Additionally, a separate element in a NFV MANO architecture is the VNF Manager, which is responsible for lifecycle management of the particular VNF. The VNFM component might be either generic, treating the VNF as a black box and performing similar operations for various VNFs, or there might be a vendor-specific VNFM that has unique capabilities for management of a given VNF. Both VIM and VNFM communication are performed via appropriate reference points, as defined by the NFV MANO architecture.
The OSM project was initially considered a multi-VIM platform, and at the time of this writing, it supports OpenStack, Vmware and OpenVIM. OpenVIM is a lightweight VIM implementation that is effectively a python wrapper around libvirt and a basic host networking configuration. At the time of this writing, the OSM VCA has limited capabilities, but still can be considered a generic VNFM based on JuJu Charms. Further, it is possible to introduce support for vendor-specific VNFMs, but additional development and integration efforts might be required on the Service Orchestrator (Rift.io) side.
Release 1 of the OPEN-O project supports only OpenStack as a VIM. This support is available as a Java-based driver for the NFVO component. For further releases, support for VMware as a VIM is planned. The Open-O Rel.1 platform has a generic VNFM that is based on JuJu Charms. Furthermore, the pluggable architecture of the OPEN-O platform can support any vendor-specific VNFM, but additional development and integration efforts will be required.
At the time of this writing the reference implementation of the CORD platform is architectured around OpenStack as a platform to spawn NFV workloads. While there is no direct relationship to the NFV MANO architecture, the XOS orchestrator is responsible for VNF lifecycle management, and thus might be thought of as the entity that provides VNFM-like functions.
When Cloudify was adapted for the NFV use case, it inherited plugins for OpenStack, VMware, Azure and others that were already available for general-purpose cloud deployments. So we can say that Cloudify has MultiVIM support and any arbitrary VIM support may be added via the appropriate plugin. Following Gigaspaces’ reference model for NFV, there is a generic VNFM that can be used with a Cloudify NFV orchestrator out of the box. Additional vendor-specific VNFM can be onboarded, but appropriate plugin development is required.
Capabilities to provision end-to-end service
NFV service provisioning consists of multiple steps, such as VNF instantiation, configuration, underlay network provisioning, and so on. Moreover, an NFV service might span multiple clouds and geographical locations. This kind of architecture requires complex workflow management by an NFV Orchestrator, and coordination and synchronisation between infrastructure entities. This section provides an overview of the various orchestrators' abilities to provision end-to-end service.
The OSM orchestration platform supports NFV service deployment spanning multiple VIMs. In particular, the OSM RO component (openmano) stores information about all VIMs available for deployment, while the Service Orchestrator can use this information during the NSD instantiation process. Meanwhile, underlay networking between VIMs should be preconfigured. There are plans to enable End-to-End network provisioning in future, but OSM Rel. 1 has no such capability.
By design, the OPEN-O platform considers both NFV and SDN infrastructure regions that might be used to provision end-to-end service. So technically, you can say that Multisite NFV service can be provisioned by OPEN-O platform. However, the OPEN-O Rel.1 platform implements just a couple of specific use cases, and at the time of this writing, you can't use it to provision an arbitrary Multisite NFV service.
The reference implementation of the CORD platform defines the provisioning of a service over a defined CORD Pod. To enable Multisite NFV Service instantiation, an additional orchestration level on the top of CORD/XOS is required. So from this perspective, at the time of this writing, CORD is not capable of instantiating a Multisite NFV service.
As Cloudify originally supported application deployment over multiple IaaS providers, technically it is possible to create a blueprint to deploy an NFV service that spans across multiple VIMs. However underlay network provisioning might require specific plugin development.
Interaction with standardization bodies and relevant communities
Each of the reviewed projects has strong industry community support. Depending on the nature of each community and the priorities of the project, there is a different focus on collaboration with an industry, other open source projects and standardization bodies.
Being hosted by ETSI, the OSM project closely collaborates with the ETSI NFV working group and follows the appropriate specifications, reference points and interfaces. At the time of this writing there are no collaborations between OSM in the scope of the OPNFV project, but it is under consideration by the OSM community. The same relates to other relevant open source projects, such as OpenStack and OpenDaylight; these projects are used “AS-IS” by the OSM platform without cross collaboration.
The OPEN-O project aims to integrate both SDN and NFV solutions to provide end-to-end service, so there is formal communication to the ETSI NFV group, while the project itself doesn’t strictly follows interfaces defined by the ETSI NFV IFA working group. On other hand there is strong integration effort with the OPNFV community via initiation of the OPERA project, which aims to integrate the OPEN-O platform as a MANO orchestrator for the OPNFV platform. Additionally there is strong interaction between OPEN-O and MEF as a part of the OpenLSO platform, and the ONOS project towards seamless integration and enabling end-to-end SDN Orchestration.
Having originated at the On.LAB (recently merged with ONF) this project follows the approach and technology stack defined by ONF. As of the time of this writing, the CORD project has no formal presence in OPNFV. Meanwhile, there is communication with MEF and ONF towards requirements gathering and use cases for the CORD project. In particular, MEF explicitly refers to E-CORD and its applicability for defining their OpenCS MEF project.
While the Cloudify platform is an open source product, it is mostly developed by a single company, thus the overall roadmap and community strategy is defined by Gigaspaces. This also relates to any collaboration with standardisation bodies: GigaSpaces participates in ETSI-approved NFV PoCs where Cloudify is used as a service orchestrator, and in an MEF-initiated LSO Proof of Concept, where Cloudify is used to provision E-Line EVPL service, and so on. Additionally, the Cloudify platform is used separately by the OPNFV community in the FuncTest project for vIMS test cases, but this mostly relates to Cloudify use cases, rather than vendor-initiated community collaboration.
Summarising the current state of the NFV orchestration platforms, we may conclude the following:The OSM platform is already suitable for evaluation purposes, and has relatively simple and straightforward architecture. Several sample NSDs and VNFDs are available for evaluation in the public gerrit repo. As a result, the platform can be easily installed and integrated with an appropriate VIM to evaluate basic NFV capabilities, trial use cases and PoCs. The project is relatively young, however, and a number of features still require development and will be available in upcoming releases. Furthermore, lack of support for end-to-end NFV service provisioning across multiple regions, including underlay network provisioning, should be considered in relation to your desired use case. Considering mature OSM community and close interaction with ETSI NFV group this project might emerge as a viable option for production-grade NFV Orchestration. At the time of this writing, the main visible benefit of the OPEN-O platform is the flexible and extendable microservices-based architecture. The OPEN-O approach considers End-to-End service provisioning spanning multiple SDN and NFV regions from the very beginning. Additionally, the OPEN-O project actively collaborates with the OPNFV community toward tight integration of the Orchestrator with OPNFV platform. Unfortunately, at the time of this writing, the OPEN-O platform requires further development to be capable of providing arbitrary NFV service provisioning. Additionally a lack of documentation makes it hard to understand the microservice logic and the interaction workflow. Meanwhile, the recent OPEN-O and ECOMP merge under the ONAP project creates powerful open source community with strong industry support, which may reshape the overall NFV orchestration market.The CORD project is the right option when OpenFlow and whiteboxes are the primary option for computing and networking infrastructure. The platform considers multiple use cases, and a large community is involved in platform development. Meanwhile, at the time of this writing, the CORD platform is a relatively “niche” solution around OpenFlow and related technologies pushed to the market by ONF.Gigaspaces Cloudify is a platform that already has a relatively long history, and at the time of this writing emerges as the most mature orchestration solution among the reviewed platforms. While the NFV use case for a Cloudify platform wasn’t originally considered, Cloudify's pluggable and extendable architecture and embedded workflow engine enables arbitrary NFV service provisioning. However, if you do consider Cloudify as an orchestration engine, be sure to consider the risk of having the decision-making process regarding the overall platform strategy controlled solely by Gigaspaces.