Network functions virtualization (NFV) has reached a fork in the road. It is widely accepted by operators as the way forward for their future networks (two-thirds of mobile operators expect to deploy it somewhere in their systems by 2020, according to Rethink Technology Research’s most recent survey of 76 MNOs), But those timelines could be pushed back significantly if it falls victim to fragmentation, as is looking highly likely at various places in the stack, from management and orchestration (MANO) to NFV infrastructure (NFVi).
As with any situation where a promising technology threatens to fail through incompatibility, there is a chance for an organization to have significant market impact if it can provide a common solution to put things back on a unified track. In MANO, the AT&T-driven ONAP (Open Network Automation Protocol) initiative aims to perform that service, if it can get the other major contender for a de facto standard, ETSI’s Open Source MANO (OSM) to work towards convergence,
At the infrastructure level, Intel has spotted the chance to solidify its powerful influence over virtualization infrastructure, while its close partner Nokia has unveiled AirGile, a technology which allow any virtual network function (VNF) to share a common database, allowing them to move between any type of underlying NFVi.
Both these developments were showcased at Intel’s Network Builders’ Summit, a one-day event held alongside LightReading’s NFV and Carrier SDN conference. The Summit took place against a background of disillusionment with NFV, which was acknowledged even by Intel – which sees NFV and software-defined networking (SDN) as its golden ticket into carrier networks, one those systems are converted into software running on standard white box hardware (built on Intel processors, of course). But Chandresh Ruparel, director of ecosystem strategy for Intel Network Builders, accepted there was “discouragement” about the pace of progress in NFV.
One of the major issues is fragmentation in NFVi. He told the audience: “There are too many NFVi configurations in the industry. The amount of time, especially with newer technologies, to just put together this base foundation is months. That is the time the community is not spending on innovation, on value-add. This is time spent on cost for development and lost opportunity because time to development is longer.”
As in so many areas of NFV/SDN, carriers and their suppliers need to learn lessons from the cloud world, where there are only a few NFVi configurations, compared to “tens” of options in the telecoms sector. “This has a huge impact across the board,” said Ruparel.
Of course, Intel is proferring a solution which would help shape telecoms transformation in its own image. Earlier in the summer, it introduced Select Solutions for NFVi, which provides a pre-verified bundle of options to support specific telecoms workloads, reducing time to deploy and the need to evaluate different hardware/software combinations.
The aim is to identify the requirements which are held in common between many telecoms workloads, however diverse, in order to reduce the NFVi choices and avoid fragmentation.
Despite Ruparel’s comments, this is a tougher task in telecoms than in IT/cloud environments. Of course, the ultimate goal is that a communications network will become a distributed cloud platfom, but that will entail a great deal of work to make very complex and demanding functions run in software. the initial test cycle for Select Solutions, for instance, Intel found “600 issues from an interoperability perspective”, he said.
This is because of the complexity of many workloads in this market. Even Intel has moved away, over the years, from the idea that all of these functions could run on x86-based COTS servers alone, and still deliver the same performance levels as they do with dedicated hardware. This was one motivation for its acquisition of FPGA (field programmable gate array) supplier Altera, allowing it to combine its own central processors with more flexible FPGAs, as well as DSPs (digital signal processors), to offload very demanding tasks, such as baseband processing, from the CPUs.
As well as compute power, each workload requires its own level of memory and storage. The industry is still trying to figure out the needs of each workload, said Heavy Reading senior analyst Roz Roseboro, who said: “In an IT environment, there are different workloads but not as much diversity in what they require. Telco workloads are so much more complex than the typical IT workload, and there is such a diversity of workloads that the industry is still trying to figure out, for each workload, how much memory, compute and storage is needed.”
That process can result in delays to implementation, and then a large number of incompatible configurations, hence Select Solutions, which hopes to result in a small set of core options. Intel has worked with Network Builders, a group of 267 partners, to develop the offering, in the hope of speeding up the process and helping to take the brakes off NFV adoption.
The work has resulted in a Select Fast Track Kit for NFVi, due out this month, which will provide pre-integrated hardware and software configurations based on Intel’s Xeon processors. The most recent release of Xeon silicon was particularly targeted at highly scalable and flexible platforms to support NFV, as well as artificial intelligence and blockchain, and other highly complex technologies.
The solutions, Intel says, will be complementary with those being developed by OPNFV (the Open Platform for NFV), has agreed to create testing and verification mechanisms to reduce time to deployment by saving individual operators from doing the work themselves. However, it will remain focused on infrastructure, rather than overlapping with work being done by the OpenStack community and others higher up the stack.
Meanwhile, Nokia wants operators to be able to develop VNFs and applications, and move them to any part of the network and any cloud regardless of the underlying stack. The company’s new AirGile system will support several important steps towards making the telco environment cloud-native (designed specifically for the cloud rather than just migrating existing functions, resulting in full programmability and open orchestration).
One of the core characteristics of cloud-native is statelessness – virtual machines and their newer counterparts, containers and Amazon’s LAMDAs, are called up dynamically when demand is high, and contain business logic, but do not know about the state of the network. This adds greatly to flexibility and responsiveness, compared to non-cloud-native systems in which VMs also have information about network issues such as session status, policies and subscriber information.
Nokia’s platform, according to Sandro Tavares, head of core cloud marketing, offers a Shared Data Layer which stores all that information on the state of the network, and can be shared by any VNF, enabling it to be stateless itself. This makes it easier and quicker for cloud applications to call up new VMs or containers as required.
“The VNFs from now on are only going to have the business logic. They are stateless machines,” Tavares said, an important step to a cloud-native environment for operators.
The Shared Data Layer is a modified version of Nokia’s Subscriber Data Management database, distributed between many locations, though all the VNFs use the same logical connection to reach it.
This is just one element of AirGile, which is basically a mash-up of NFV and cloud developments from various parts of Nokia, including some acquired with Alcatel-Lucent. These will gradually be integrated into a unified cloud-native platform. Among its constituent parts are a virtualized VoLTE platform and CloudBand, Alcatel-Lucent’s MANO system for telco cloud.
Nokia says operators and other partners will contribute their own developments to AirGile, and it has created a portal with a set of APIs (application programming interfaces) and a developer sandbox, to make it easier for third parties to enrich the platform. Nokia itself is contributing the interface between the VNFs and the Shared Data Layer, hoping AirGile will attract third party VNFs. This would move beyond NFVi neutrality and provide a contender for a de facto standard for shared VNF databases, which will be a critical issue for telcos as they move towards cloud-native environments, while still retaining the link between functions and the subscriber’s network experience.
While Nokia dreams of a pure cloud-native platform in which applications can move across any kind of NFV infrastructure, in reality, most operators will be dealing with legacy, physical infrastructure for many years to come, and the interworking of the new and the old is another potential area of fragmentation. TM Forum has announced a blueprint to provide a standardized way for operators to manage legacy and virtualized infrastructure simultaneously in a hybrid environment.
This blueprint is based around a standard interface designed to allow physical and virtual network elements, from multiple vendors, to interoperate. Such interfaces will be fundamental to the vision of a virtualized carrier network which not only frees the operator from the economics of a rigid, expensive hardware-based platform, but also from end-to-end vendor lock-in. Several Forums are working on such interfaces for different parts of the system – Small Cell Forum’s nFAPI, for instance, interlinks multivendor cells and controllers in a virtualized cluster, and is now part of the open source CORD initiative.
Other interfaces are less promising, however, perhaps because they are driven by the major vendors – a notorious example is CPRI, the link between remote radio heads and basebands, which is implemented differently by each supplier. Many observers were speculating that eCPRI, its new iteration for 5G, would not only be more economical and flexible for Cloud-RAN fronthaul (linking radio units with virtualized, centralized basebands), but would be fully standardized, or even open source. However, according to John Baker of software-driven networking company Mavenir, eCPRI is suffering from the same fragmentation as its predecessor.
TM Forum is determined that its new interface will not fall into that trap, and that it will have a rich accompanying set of tools and processes to make it fully usable. Its ‘Implementation and Deployment Blueprints for Hybrid Environments’ aims to provide a standardized and interoperable approach to an issue which has been given far too little attention, according to CEO Nik Willetts.
He told TelecomTV: “One important thing to discount is the idea that we’re going to wake up next week and suddenly we’re going to have a virtualized architecture. It’s going to be a hybrid world for some time to come.”
The blueprint is part of TM Forum’s well-established ZOOM (Zero-touch Orchestration, Operations and Management) project. It includes open APIs, information models, best practices and deployment guides, all surrounding the core interface. It claims to provide some management capabilities which are not currently available together in a standardized package, including Resource Function Activation and Configuration, Inventory Management, Catalog Management, and Assurance.
Willetts said: “The objective is to bring together all the TM Forum work that we’ve done on zero touch and orchestration management as well as the APIs, the information model and the architecture work, with the idea that it can sit on top of hybrid (virtualized and traditional) infrastructure. Then we can turn our attention to how that interfaces to a new generation of OSS/BSS systems.”
The impact of virtualization on operators’ business and network performance will be greatly reduced if the OSS/BSS is not also completely reworked. Willetts added: “We have to keep reminding ourselves that virtualization isn’t the goal. It’s the agility it brings that’s the goal.”
One of the operators which is most heavily involved in open initiatives for NFV/SDN, the UK’s BT, is supporting TM Forum’s latest work. Milind Bhagwat, an enterprise architect at the telco, praised the Forum’s “intent-based, catalog-driven and policy-enabled approach” to management, adding: “Operators today are faced with the massive challenge of managing hybrid infrastructure composed of virtual and physical components … The Blueprint allows us to effectively automate end-to-end service provisioning and management. And we all know, automation is the key that unlocks the benefits of network virtualization.”