When we come to write our ‘predictions for 2014’ pieces at the end of the year, Carrier SDN (software defined networking) is sure to be at the top of many lists. The dream is a compelling one for carriers trying to achieve more flexible, cost effective capacity, while breaking away from the proprietary, over-complex networks of the past.
Perhaps that could all be delivered by virtualizing the functions of the base station on standard hardware ‘ as in Cloud-RAN ‘ and in turn treating both that C-RAN and the packet core merely as two applications among many on the server. SDN platforms are starting to be adapted for telcos by companies like Intel, Cisco and Hewlett-Packard, while the traditional mobile infrastructure vendors try to maintain some control via efforts like NFV (Network Functions Virtualization). But even HP, launching its SDN applications store this week, admitted that extending SDN from the servers right out across the wide area network to end users would not be a reality until 2020 ‘ and that is a very over-optimistic estimate outside a few countries where carriers have the motivation, funds and ecosystem control to make it happen (China, in particular).
Elsewhere, a more typical timeline would be: 2014, continue to focus on conventional data build-outs (especially 3G+ expansion); 2015-16, start to roll out a small cell layer and some LTE-Advanced features; 2017-2020, introduce elements of C-RAN in a hybrid macro layer architecture and address HetNet integration; 2025, implement true SDN.
Nevertheless, as in all major architectural shifts, there will be a few frontrunners with deep pockets to justify the vendors’ enthusiasm in the early years, and while there will be no ‘big bang’ shift to SDN, the compromise and hybrid solutions which some carriers implement along the way will still generate new innovations and industry revenues. So companies like Cisco, Intel and HP hope, anyway, because for them this is about finding a really strategic place in the telco market. All of these, and other enterprise majors like Oracle, already have strong positions with carriers ‘ but in backbone infrastructure and/or the IT department. Now they want to steal the driving seat role from the mobile RAN suppliers by insinuating their IT-oriented technologies into the heart of the network. The vision of SDN spreading outwards from the enterprise, to become an end-to-end architecture underpinning the whole network-enabled population, is a powerful way to do this, even if it takes a decade, and the players are starting to lay out their roadmaps already.
Intel unveiled its SDN strategy, with a heavy nod to telcos, earlier this year, as did Juniper. The past couple of weeks have seen the turn of Cisco (which has and HP, and they will have been encouraged in their efforts by announcements from AT&T, which is increasingly joining the exclusive group of operators, mainly Asian, which are really cutting edge on the network side. It set out a revised supplier strategy which pointed firmly to its plans for a software-driven, all-IP network in future.
Despite HP’s admission that end-to-end SDN over the WAN would be a long time coming, it still believes its latest developments will help quicken the pace among carriers as well as enterprises, removing some of the current barriers. Unveiling its SDN Developer Kit (SDK) and SDN App Store, together with a suite of services and certifications. Kash Shaikh, senior director of product marketing for HP Networking, said: ‘We’re really moving the market forward while removing the barriers to software defined networking.’
With this approach, HP is trying to put itself at the center of a broad ecosystem by making programmable networks ‘programmable’ in the same way as any other platforms, with standard tools that have a wide base of developers and APIs. To demonstrate the potential for that PC-style community, it announced partnerships with key allies such as Intel, Microsoft, VMware and F5.
This sees HP trying to push SDN as a way to simplify networking by reducing the need for specialized hardware and skills, and by implementing all functions in the same standardized way. This is an important goal of the Open Networking Foundation, which HP supports heavily and which is responsible for the OpenFlow SDN network management and switching specifications. Last week, the ONF announced an open source tool specifically designed to help operators measure the complexity of their networks, which the tool’s co-creator, Stu Bailey, CTO of Infoblox, believes is a bigger barrier to growth of network services than the usual culprit, lack of bandwidth.
Bailey said: ‘For the first time in the history of the networking industry, it is becoming clear that complexity rather than bandwidth is the barrier to network growth. Today, discussions about network complexity focus on the tangle of wires and boxes, rather than the relationship of business processes to an increasingly large, dynamic, and shared global IT infrastructure.’ Developed with Professor Robert Grossman of the University of Chicago, Tapestry uses key NFV and SDN techniques and generates a ‘Network Complexity Index’ (NCI) number based on endpoint interaction data from network-wide control systems such as the DNS.
The aim of the tool is to encourage enterprises and carriers to take a new perspective on their networks by focusing their attention on SDN as a simplifying mechanism rather than a daunting architectural transformation. For instance, a hoped-for outcome of SDN would be to make it far easier to deploy new services and applications, without needing to refer to the network, an issue which is particularly important in the still-esoteric world of mobile RAN.
Rohit Mehra, VP of network infrastructure at HP, said in a statement: The Catch-22 is that to innovate through applications requires a large investment in infrastructure to develop, which becomes prohibitive. The advent of an SDN App Store, together with an SDN Developer Kit, makes this an accessible alternative for developers.’ These products will appear between November and mid-2014.
The company, working with VMware, will develop a federated system that will provide companies with unified visibility into ‘ and automation of ‘ their physical and virtual networks. This will be targeted first at the data center, and even in that environment will not appear commercially until the second half of 2014, and will then be extended to carriers.
Some operators are already showing strong interest in HP’s approach – NTT is one of the new ecosystem partners, and Telefonica is testing the new OpenFlow-supporting routers which HP launched to accompany its latest SDN offerings. The Spanish giant is evaluating how a virtual router could be deployed for its enterprise customers in future.
Cisco will make its own SDN announcements next month together with Insieme, the start-up it has funded in this area. Indeed, the Cisco/Insieme cooperation is expected to result in an ambitious attempt to redefine the SDN sector before it has even got started. The partners will next month release their Application Centric Infrastructure (ACI), which Network World rightly describes as ‘hardware defined networking’ rather than pure SDN, because it integrates the physical and the virtual, rather than moving away from the former altogether.
This can be seen as a way for Cisco to preserve some of its specialized hardware revenues as well as its overall ability to drive enterprise networking patterns and platforms. The vendor itself portrays it as an interpretation of SDN that will be more practical and robust in the real world in many cases, and will go beyond what Frank d’Agostino, senior director of technical marketing and solutions engineering at Insieme, calls ‘a software overlay on a shim trying to solve a problem that’s three years old’. That approach, which he claims is the one adopted by many rivals, would force enterprises and carriers to ‘support all SDN models if you go that route’, at a time when there are scores of immature and competing models to choose from.
Talking at last week’s Interop show in New York about ACI – which is being developed and productized, as the Nexus 9000 line, by Cisco’s 80%-owned spin-in Insieme – he said it will focus on programming the physical infrastructure-based service as well as the hypervisor-based virtual environment. ‘We want to program the physical service the same as a virtual service,’ he explained, taking clear potshots at VMware and others with a ‘software-only overlay’ strategy. D’Agostino claims this approach risks loss of visibility between the application service and the network; limited scalability; lock-in by a particular hypervisor; focusing down on a ‘small subset’ of programmability applicable only to ‘one-eighth’ of the total; and pricing on a per-VM basis. ‘If you’re paying per VM for network virtualization, you’re wasting your company’s money,’ he insisted.
Meanwhile, ACI combines software with merchant silicon and custom ASICs in hardware that is optimized for availability of services, whether virtual or not, and provides an automated common operational model through ‘unified’ APIs, which will be published and will not require an SDK (a dig at HP’s new SDK and the bid by some vendors to lock customers in via developer tools). ACI promises to support any application, hypervisor and physical edge with the visibility of hardware. ‘We don’t want the application to be network aware; we want the network to be application aware,’ said d’Agostino, revealing all his Cisco roots.
Before Cisco really tries to harness SDN for the long-standing ‘network is the computer’ vision, it has already unveiled another contribution to the software-defined jigsaw, its Network Convergence System (NCS). This is the vendor’s big play to provide a fabric to support virtualization and programmable networks ‘ extending its fabric play in the enterprise out to the WAN. In doing so, it is taking on Juniper’s PTX and Alcatel-Lucent’s 7950 XRS routers ‘ while in theory, SDN might eventually run routing functions entirely on x86, for now, the router suppliers are making hay out of SDN-enabled versions of their products. Earlier this year, as part of its ONE (Open Network Environment) initiative for programmable networks, Cisco updated its edge routers and switches, and now it has turned to the high end.
It describes NCS as “a network fabric family designed to serve as the foundation of a massively scalable, smarter and more adaptable internet.” Its programmability and virtualization capabilities are designed to enable operators to accelerate the move to SDN and NFV. The family has three members, NCS 6000, 4000 and 2000, which can be managed as a single integrated system. The 6000 is an IP/MPLS routing platform capable of 5Tbps per slot and 1.2 petabits per second per system; the 4000 supports OTN, WDM, Sonet and Ethernet applications at 400Gbps per slot and 6.4Tbps per system. The 2000 connects DWDM transport networks at 100Gbps and above.
“When deployed as part of the Cisco One Service Provider Architecture, which embraces and extends SDN/NFV, the Cisco NCS family can help network operators reduce total cost of ownership by 45% while consuming 60% less power,” the company said, pointing to the main reasons why cellcos will take programmability seriously in the short term ‘ less for innovative new services, more to reduce cost and power consumption by allocating capacity more dynamically, and sidelining proprietary kit.
The fabric harnesses a range of previous Cisco developments which aim to adapt its core products to the world of mobile broadband, M2M and cloud. As well as the existing SDN innovations for the data center, such as Cisco’s Unified Computing System and Dynamic Fabric Automation, it uses the nPower X1 processor, a 400Gbps monster which the vendor unveiled a few weeks ago. This has been particularly targeted at networks that need to cope with the coming traffic explosion sparked by the internet of things. As well as being the most integrated ASIC in Cisco’s history, it is also the firm’s most programmable processor. It supports eight times the throughput at 25% of the power consumption compared with Cisco’s previous network processor, and combines all packet processing, traffic management and I/O functions.
KDDI, Telstra and BSkyB (Sky) are the first public customers for NCS and Mike Wright, the Australian incumbent’s executive director of networks, commented: “This network evolution will help us to continue to have Australia’s largest and most reliable IP and 3G/4G wireless network.’
The idea that deployment of an all-IP network should be the opportunity to introduce an entirely new, efficient and flexible architecture is not confined to Telstra and the well-known Asian all-IP leaders like SK Telecom or NTT. AT&T is moving into their ranks too and its latest procurement strategy positions it firmly as an SDN company.
It has announced its Supplier Domain Program 2.0 ‘ an update to the original 2009 program, which categorized all its equipment purchases into 14 categories or domains, typically with two suppliers each, in order to streamline purchasing and network integration. The second iteration is designed to ease AT&T’s path to software-driven, all-IP networking, and eventually to reduce capex and opex levels. Tim Harden, president of supply chain at the carrier, said the goal was ‘to ensure that each investment accelerates our move towards an advanced all-IP broadband, all-wireless, and all-cloud infrastructure.’
The revised five-year program, in AT&T’s own words, ‘represents a seismic shift in how AT&T will do business as the hardware and software functionality in the network will be separated to give the ability to simplify and scale the network’. There will be nine domains or categories, but unlike the previous domain system, each could have as many as seven vendors, (though some will only have two). Some of the initial set of domains will remain while others will be modified or completely changed. New ones, of course, include NFV.
The resulting new architecture will be geared to reducing time to market for new products and services, and lowering cost of roll-out, by separating hardware from software functionality, and network control plane from forwarding planes, as well as enhancing the capabilities of the software layer. However, AT&T is not a great believer in the virtualized or Cloud RAN, at least in the short term. It will look first to re-architect its core and metro networks and implement those functions in software ‘ and in this respect, will set a more common pattern among carriers than China Mobile with its new-generation RAN designs.
“We eventually see this extending to the RAN, but that is not our initial focus,’ AT&T said. “AT&T operates as an integrated carrier end to end. Therefore, the customer will be able to interact with our network in an all-IP environment for their wired as well as wireless requirements.’
The operator will start choosing its new wave of vendors late this year and award contracts through most of 2014, and said that over a five-year period its new strategy will reduce capex, though that will not be an immediate benefit ‘ the operator is not revising its capex guidance at all for 2014-2015. Its schedules are still relatively ambitious though, looking for serious impact from SDN in a far shorter timeframe than the majority of operators, especially in Europe.
The RAN may be down the track, but that could benefit the invasion of the IT vendors in the carrier network ‘ there is far greater logic to HP and Cisco pushing Ericsson aside in a virtualized core than in its RAN heartland. Not that this will be all to the benefit of the traditional router makers, despite their SDN activities. Cisco, ALU and Juniper were the Domain 1.0 IP/Ethernet choices, and all have worked hard to modernize their offerings for the software-driven environment ‘ but AT&T’s new approach could also let SDN specialists like Big Switch and Arista in the door. Nonetheless, until the shape of NFV standards is a great deal clearer, it will be the RAN vendors which need to up their game. With a host of announcements expected at this week’s Interop show in the SDN field, the ball will now be firmly in Ericsson’s court.
The flood of NFV releases from large OEMs begins:
The OEMs will be stepping up their virtualization activities as they race to offer platforms that support NFV, the SDN-related technology that, of all the options emerging, provides them with the best chance to keep the shift from specialized hardware under their control nevertheless. Ericsson and its rivals are working hard to develop NFV implementations that deliver the benefits carriers are targeting, while maintaining a high level of proprietary influence over the resulting system ‘ whether this relates to the software itself or the addition of specialized hardware for functions that ‘just cannot’ be virtualized without severe impact on network performance.
Expect streams of NFV releases in the coming year, especially at February’s Mobile World Congress, but they are already starting now, before the specifications are even frozen. The latest are:
NSN has virtualized the latest version of its NetAct operations support system (OSS) to provide off-the-shelf management of NFV. NetAct 8 is a virtualized, single OSS for managing radio, core and transport networks, which fits neatly with the vendor’s Liquid Net platform. This is increasingly running every aspect of the carrier network in the same way, and with a high dose of software to allocate resources flexibly where required and increase overall efficiency.
Virtualization enables NetAct 8 to deliver high availability independently of any single hardware platform, thus reducing capex as well as opex, the company said. “We’ve brought OSS into the cloud age, using automation and virtualization to cut running costs, add greater flexibility and ensure superior network quality. NetAct 8 covers network operations with tight link to service operations, eliminating the need for multiple management systems,” said Peter Patomella, the firm’s VP for OSS.
Jane Rygaard, head of marketing for customer experience management, core and OSS at NSN, said that virtualization enabled it to be far more flexible when specifying hardware to run NetAct, but with the new release, ‘we no longer need the dedicated hardware for the different functions’. That approach will start to extend across the packet core, with all its various dedicated functions and boxes, and Rygaard believes virtualizing the OSS is an essential first step for any operator looking to do the same for its core network. She added: ‘From the OSS side, it has been extremely important for us to say that we know how to be ready for that next step when the network side is ready to virtualize.’
Alcatel-Lucent is opening a cloud R&D facility in Israel to bolster its NFV platform. This is the first concrete move in the R&D aspect of its corporate ‘Shift Plan’, which aims to reduce costs and focus R&D on a smaller number of critical areas, notably IP networks and ‘ultra-broadband access’.
As part of its NFV push, ALU is also creating an ecosystem program to support its CloudBand Ecosystem Program, which it describes as the first open community of service providers and vendors adopting NFV. So far, it has signed 15 partners, including Deutsche Telekom, Telefonica and Intel.