It’s a rare statement in an industry driven by hype and over-expectation, but virtualization and software-defined networking (SDN) have become a reality for mobile carriers in an incredibly short time.
Given the complexity of the concepts, and the disruption to operators’ traditional platforms and working practices, it is surprising to see the level of acceptance, at this early stage, that SDN and virtualization – two different but often intertwined developments – will be part of mainstream networks in the near term.
This will be a process spread over a decade and more. Some carriers are already testing systems based on the key virtualization approach, ETSI’s NFV (Network Functions Virtualization), and on SDN ideas. Commercial implementations will follow as early as 2015, but these will be by a very few carriers (such as Telefonica for NFV and AT&T for SDN), and just in selected elements of the network.
Those elements will need to have an obvious business case for implementation in software, as well as existing systems being ripe for replacement – the evolved packet core (EPC) will be an early candidate for virtualization, while the RAN will not, except in virtually greenfield cases like China Mobile.
It will take many years before the bulk of mobile operators have implemented NFV and/or SDN, and before even the leaders have applied these approaches to the bulk of their networks. But nevertheless, progress has been lightning-fast by the standards of this market, and it shows that there is a real need and business case to take a radically new approach to carrier networks.
Merely upgrading the air interface and making the hardware more efficient is no longer adequate to keep up with operators’ pressures – to support rising volumes of data while maintaining quality of service and profitability, and fending off over-the-top rivals and the threat of the dumb pipe role. These pressures will only be met, from the second half of this decade, by completing changing the cost base, the ecosystem and the service delivery. Operators need three critical elements, all of which they believe a software-led approach could deliver – a far lower cost base for RAN, core and transport; full interoperability to end vendor lock-ins and create an open, PC-style market; and the flexibility to add and remove new services, and handle big data, in a dynamic way.
Lower cost is achieved primarily by moving functionality from dedicated boxes to applications on standard hardware, and by allowing for dynamic allocation of network resources where they are needed, rather than over-provisioning every element to handle peaks in demand. Flexibility is helped by that dynamic approach, and because new instances of the virtual machines, in which the network functions now run, can be created or upgraded on the fly.
Perhaps the most challenging requirement has been interoperability, supporting multivendor networks and open ecosystems, and in future a cloud-like approach to network infrastructure. Many carriers still fear that key standards like NFV will somehow be hijacked by major suppliers, which will add proprietary ‘secret sauce’ or fail to interwork smoothly with rivals’ software. Open source initiatives are important in this respect, and the software revolution is being driven, to some extent, by the open community. The Linux Foundation, which hosts the important open SDN platforms, OpenDaylight and OpenStack, recently announced an associated virtualization effort, OPNFV (see Wireless Watch October 2 2014). This work will provide a bridge between carrier SDN and NFV, and also helps to secure NFV’s position as the de facto virtualization platform even more firmly.
Indeed, interoperability lies at the heart of the second wave of ETSI’s work on NFV, as well as other carrier virtualization collaborations round the world. Earlier this year, the ETSI NFV ISG (industry specifications group) confirmed that it would not wind up its work with the achievement of the basic specifications, but would carry on to a second phase, which would tackle issues such as interface specs for interoperability, and interaction with legacy OSS/BSS systems.
The detailed goals of this new work have been outlined in a white paper released by 30 operator members of the ISG. This is available on the group’s website in draft form and will be finalized in November. It lists interoperability, support for layered architectures, interaction with legacy OSS/BSS, and definition of new streamlined service models and processes, as the main priorities for the second round.
In a more visionary tone, the paper also envisages a “telecommunications cloud”, incorporating selected aspects of both NFV and SDN in a flexible “cloud of clouds” environment which can be scaled dynamically according to requirements. This will be built on a stacked architecture in which different network orchestrators, cloud operating systems and hypervisors can be mixed and matched.
In that layered architecture, NFV would be able to support a hierarchy of orchestrators, each of which could operate independently and support different operators, tenants or services. That shows how multivendor interoperability, in a software networks world, is not just about traditional issues of encouraging supplier competition within one carrier’s deployment. It is also about the future way that networks and their owners will operate, supporting large numbers of virtual service providers and enterprise customers, and creating large shared infrastructures supporting many separate services.
This could see the mobile market, in a decade or more’s time, looking far more like the cloud segment does now, and ‘5G’ developments will have to look in this direction, towards very broad ‘as a service’ platforms supporting many providers, rather than the current model where each MNO builds a largely separate network and then supports a select number of exclusive MVNO deals.
However, despite the importance of open source and standards groups, the detailed technical work of leading operators and their suppliers will remain essential to ensure the dream of interoperability does not languish in committee rooms.
Tests using pre-standard gear are important to see what works in the real world, and to feed into standards efforts. Progress has been particularly rapid in the optical field, according to the Optical Internetworking Forum (OIF), which says its prototype testing shows that
interoperable SDN is approaching commercialization. Working with the Open Networking Foundation (ONF), the Forum has been overseeing testing in several carrier labs in north America, Europe and China.
The first results of these tests were revealed earlier this month (http://www.oiforum.com/public/documents/OIF-ONF_Global_Transport_SDN_Demo_White_Paper_2014.pdf)
and, according to Dave Brown, OIF board member and director of optical products marketing for Alcatel-Lucent, commercialization is not far away. “We are still in the early stages, but these are exciting and positive results and very encouraging,” he told LightReading, adding: “The prototypes that were shown were pretty stable. And we were able to do what we were trying to do and identify any issue or gaps that need to be addressed by groups like the OIF or ONF” – but of 18 issues identified, none was a “showstopper”.
The application tested was cloudbursting, or increasing the bandwidth between data centers to support cloud services, a common business case for carrier SDN. Based on this kind of work, implementation agreements can be submitted to standards bodies such as ETSI, IEEE and the ITU.
Another sign of the progress made on multivendor NFV came from Japan’s NTT Docomo this week. The operator, always at the forefront of disruptive mobile network trends, said it had completed proof-of-concept (PoC) trials that verified the feasibility of NFV in a test network using software and equipment from six suppliers – Alcatel-Lucent, Cisco, Ericsson, Huawei, NEC and Nokia Networks. The trials focused on the packet core and verified the interoperability of each vendor’s EPC software with another’s platform, testing functions such as scalable data processing capability for congested data traffic, and immediate network recovery in the event of hardware failure. In most cases, the software was originally designed for dedicated appliances and has been converted to run on multipurpose hardware, but in future vendors will design such software from scratch for off-the-shelf servers and routers.
Seizo Onoe, Docomo’s CTO, said in a statement: “Thanks to collaborative efforts in the mobile network industry, NFV is soundly moving toward practical realization, rather than just ending up as a pie-in-the-sky dream. I am delighted that we have confirmed the feasibility of NFV through multivendor initiatives with top ICT players.”
While the Docomo trial focused on the big names of end-to-end networking, vendors throughout the chain are racing to virtualize their portfolios and their code bases for standard hardware. For instance, this week saw IP equipment maker Sonus announcing SWe (Software Edition) implementations of its PSX routing and policy server and of its Diameter Signal Controller (DSC). This allows for a policy engine or DSC to be deployed alongside other applications on the same hardware. Sonus had already claimed the world’s first virtualized SBC (session border controller). It has demonstrated interoperability of its virtualized products with Juniper’s Contrail SDN and NFV platform, and others will follow.
Of course, the challenge for the major network vendors is to offset the loss of margins and client control which came with dedicated, proprietary hardware. Most are trying to do this by moving their control up a level, and seeking to provide the cloud platform in which all the multivendor products are deployed and managed. In its statement supporting Docomo’s PoC effort, for instance, Ericsson said pointedly: “The Ericsson Cloud Execution Environment allows operators to deploy Ericsson’s and other vendors’ virtualized hardware and software products. This also helps them to allocate their network resources more flexibly, simplifying network configuration and ensuring savings in operational and capital expenditure.” Its rivals have similar initiatives, such as ALU’s CloudBand architecture and Cisco’s Intercloud ‘cloud of clouds’.
ALU’s latest partner in NFV and SDN trials is KT of Korea – the two companies have announced a technical collaboration agreement and will create a proof of concept based on the vendor’s virtualized EPC, CloudBand NFV and Nuage Networks SDN platforms. The alliance will seek to drive virtualization into KT’s development of its next generation ‘Giga Network’, which will be underpinned by NFV.
“Alcatel-Lucent and KT share a vision of the strategic importance of virtualization and cloud infrastructure and service formation. We look forward to helping KT transition its EPC network functions to operate in an NFV/SDN environment., said Chi Il Yoo, head of ALU Korea.
For those which fail to secure significant share in cloud platforms and virtualized systems, spending on conventional equipment will be squeezed by the SDN and NFV trends, if slowly at first – even the most software-driven operators will not replace existing systems until they are nearing end-of-life anyway, and will continue to expand and enhance those systems while they are still useful. This is particularly true in the RAN, where many operators have only recently upgraded to LTE, using traditional architectures in most cases.
However, there have been some signs of pressure on traditional network equipment capex spending, especially in the US, which has hurt recent quarters for vendors like Juniper. A recent research note from analysts at Jefferies indicate that the overall capex spend among north American wireless carriers has been “soft” this year, despite aggressive spending by Verizon on LTE capacity and DAS systems. AT&T has been the most sluggish and many have pointed at its ambitious plan to shift to an SDN architecture in many areas of its network, starting from 2015. It has announced its ‘Domain 2.0’ plan to implement ‘elastic’ networks and services via its ‘User-Defined Cloud’ program, and to introduce SDN in key network elements. That will see the operator develop a homegrown SDN controller and introduce new suppliers to its inner circle alongside the traditional names – so far, its Domain 2.0 list includes Ericsson, ALU, Fujitsu, Amdocs and Juniper, but also smaller specialists like Metaswitch, Tail-f and Affirmed Networks. Having been left out in the cold, Cisco quickly moved to acquire Tail-f.
However, there are other reasons for AT&T’s relatively low spending this year, according to Jefferies. It has slowed down on DAS deployment, unlike Verizon; and has been implementing fewer new cell sites, concentrating instead on upgrading existing ones. And vendor selections are only just starting for another important program, ‘Turf 3.0’, which covers the professional services organizations responsible for installing wireless equipment on towers. The vendor is consolidating from 27 to 15 turf vendors but the new deals will only kick in next year.
But the big change will come with SDN, and last month AT&T said it had tested its User Defined Network Cloud platform in the University of Texas at Austin, and it was ready to be commercially deployed in that city by year end, and in other cities during 2015. “What the cloud did for the management of data centers, Network on Demand will do for corporate networks,” wrote AT&T’s Roman Pacewicz on a blog. “This is a breakthrough on-demand service that will help companies easily add or change services on their own, in near-real time.”
In an earlier Domain 2.0 white paper, AT&T wrote: “AT&T services will increasingly become cloud-centric workloads. Starting in data centers and at the network edges, networking services, capabilities, and business policies will be instantiated as needed over the aforementioned common infrastructure. This will be embodied by orchestrating software instances that can be composed to perform similar tasks at various scale and reliability using techniques typical of cloud software architecture.”
While AT&T is one of the most vocal supporters of the new approach – as it has been for other ways of disrupting the network ecosystem, such as small cells – it is certainly not a lone voice. Other operators will be watching its progress, and the rising tide of carrier PoCs for various elements of NFV and SDN. Vendors will need to change their way of thinking at the deepest level, if they are to survive this slow but fundamental change in their markets.
SDN and NFV:
Virtualization implements some or all network functions as virtual machines running on off-the-shelf hardware, saving on the cost of ownership of a string of dedicated one-function boxes. It also creates tunnels between those VMs so they can be reconfigured on the fly without affecting the whole system, adding to flexibility and allowing new instances of VMs to be deployed or taken down rapidly.
SDN can be implemented entirely separately, but has specific benefits for virtualization, such as the ability to manage huge numbers of VMs from the cloud. SDN separates the control and data planes, centralizing all the control of the huge number of VMs machines on standard servers and allowing the behavior of the network to be freely programmable, using standard interfaces and common developer tools.
Among mobile operators, virtualization is widely seen as a nearer term strategy, at least for certain parts of the network such as the packet core, than full-blown SDN. However, general consensus is that the full benefits of virtualization will only be achieved as part of an SDN transition, even if that takes 10 years to achieve.
The ETSI set of specifications for network function virtualization, NFV, has been an important factor behind operators’ unexpectedly intense interest in near-term deployments – giving them the confidence of a widely supported set of specs which have been carrier-driven and tailored for telcos’ specific needs.
NFV wave two:
The carrier-dominated ETSI NFV ISG will release a set of documents in January to start to address the second phase of requirements which it has identified for its platform, and which it has outlined in a new white paper.
The January release will include:
• NFV Infrastructure and its components
• Management and orchestration (MANO)
• Software architecture
• Reliability and availability
• Performance and portability
• Security
The ISG also aims to define interfaces between key components of the NFV ecosystem to encourage interoperability and multivendor systems. These will include;
• Orchestrator and Virtual Infrastructure Manager (VIM)
• Orchestrator and the Virtual Network Function Manager (VNFM)
• VNFM and Virtual Network Function (VNF)
• VNF and NFV Infrastructure (NFVI)