The initial excitement around Cloud-RAN, which bubbled up when China Mobile hurled its disruptive concept at the industry five years ago, has largely worn off. SK Telecom, Orange and T-Mobile all expressed scepticism about its advantages at the recent Mobile World Congress, at least in the short to medium term.
Initially, operators were excited at the prospect of slashing the cost of ownership of new networks by sharing resources flexibly among cell sites and replacing proprietary, specialized hardware with white boxes running virtual network functions (VNFs). But of course, C-RAN proved far harder than most had hoped. For operators which saw it mainly as a way to slash operating expenditure, the costs of VNFs and white boxes remain unclear and the need for high quality fiber for fronthaul looms large. For those which saw C-RAN as a way to support new revenues via an agile network, the business cases are far from accepted either.
In a new report from Rethink Technology Research, 74 mobile and converged operators worldwide were surveyed about their plans for centralized and/or virtualized RANs (vRANs). The report also highlights the concerns and uncertainties which have led to most MNOs pushing back their timelines for vRAN.
However, while progress has been slower than expected at the time of Rethink’s last report on this topic, 18 months ago, there is still a high belief that vRAN is essential to make full economic sense of 5G. For the majority of MNOs, 5G is both a driver and enabler of vRAN, and one reason for delaying large-scale vRAN deployments is because these will make more sense when accompanied by other technologies that are only just emerging:
• 5G New Radio with its support for a sliceable network
• standards for management and orchestration (MANO) in vRANs
• edge compute architectures for mobile networks
• more flexible and affordable fronthaul options based on Ethernet.
These issues are being addressed by various bodies:
• 3GPP for NR, slicing and the functional splits between virtual and physical network functions
• ETSI, Facebooks’ Telecom Infra Project, the new ORAN initiative, and the Open Networking Foundation all studying approaches for MANO
• ETSI, TIP and OpenFog are among the bodies focused on edge compute for telcos
• IEEE and the CPRI group are both defining standards for radio-over-Ethernet fronthaul
But waiting for all these initiatives to complete their work, and to find common cause to drive an interoperable platform, is delaying real world deployment. Rethink’s forecast – based on its operator survey and a broader survey of vendors, regulators and other stakeholders – is that vRAN deployments will only overtake traditional, in new sites, in 2022, largely driven by 5G. And as the Figure shows, it will be 2024 before more than half of Tier 1 and 2 MNOs have large-scale vRAN deployments. Before then, the rising numbers of vRAN sites will be driven by a relatively small group of big deployers, mainly in east Asia and North America, and by localized vRANs serving cities or enterprises and based around small cells or microcells.
The caution about vRAN suggests that most operator will wait until they have an urgent need for a new architecture – usually to make the case for 5G, to transform their cost models and their range of services. With a few exceptions, that wide-scale 5G roll-out will come towards the middle of the next decade, and before that, most MNOs will be deploying 5G in a conventional way, to add capacity to 4G mobile broadband services or support fixed wireless.
T-Mobile USA’s CTO, Neville Ray, expressed this lack of immediate urgency in comments at MWC, when he said: “I think that vRAN and ORAN are exciting opportunities, but I don’t see that they’re going to materialize in the near term.” AT&T had announced ORAN – the merger of its XRAN inhouse initiative, which it recently open sourced under the auspices of the Linux Foundation, with the Cloud-RAN Alliance. This gives greater scale and profile to the effort to define a carrier-led architecture for vRAN, one which will enable operators to open up their supply chains and embrace open technologies and non-traditional vendors. But it still has a long way to go to prove its business case to a wide base of operators, especially those which, unlike AT&T, do not have the skills or appetite to be technology pioneers.
Ray continued: “I think that virtualized RAN is clearly an exciting opportunity, it’s just early. On ORAN, if you think about the necessary R&D and investment into bringing what is very complex technology to market, I think there’s a period of time where those investments need to be realized to support the level of investment that’s needed to bring the technology to market in the first place.”
In the meantime, established vendors are doing a lot of the heavy lifting, however much MNOs would like to be less reliant on them. He said: “Obviously we’re evaluating capabilities out there but we’re spending a lot of time with our existing vendors on making sure we can deliver products in the timeframe that is required and needed.”
Orange’s director of spectrum strategy and planning, Yves Bellego, was also urging caution in Barcelona, despite strong technical results from the French telco’s vRAN trials, which are some of the longest-running in the industry outside east Asia. However, they have so far failed to prove significant cost or business benefits, he said.
“We did not go beyond trials or a proof of concept because of the business case,” he acknowledged. “As of today it remains cheaper and simpler to upgrade each and every site.”
While Orange was an early exponent of virtualization in general, and vRAN in particular, it made it very clear, a couple of years ago, that it would no longer be satisfied with justifying these architectural changes on the basis of cost alone. It would also insist on seeing new revenue streams or system capabilities. Bellego said: “We do not think savings is the initial driver. There are initial costs we must take care of and so there must be other drivers. We have been looking to see if there are additional capabilities on radio. Could we improve capacity by having pooled functions? Pooling BBUs [baseband units] could help with automation and edge computing.”
In other words, the vRAN must be considered in tandem with the entire network platform, including the edge (see separate item). The work by TIP and ORAN to make fronthaul and white boxes both powerful and affordable will help make the case, but will not be enough on its own.
It is a very complex process to move from a traditional to a virtualized RAN. Unlike virtualizing the packet core, for instance, which is a discrete system, it affects the biggest aspect of capex and opex in the network, and all the internal processes which are bound up with the RAN. SK Telecom of South Korea, an early and advanced exponent of vRAN architectures, was nevertheless sounding a cautionary note at MWC. It said that some of the main barriers had been internal, particularly from its operations department, concerned about added complexity and even security risks.
Like other operators, such as the UK’s BT, SK Telecom said its operations teams had warned of the complexity of implementing open hardware and software. This can be unpredictable and of varying quality – and risk vulnerabilities – because of the open development process. To make it secure and fully carrier-grade, a high degree of inhour or consultancy work needs to be done on implementation and optimization.
Park Jin-hyo, SK Telecom’s CTO, told LightReading that his firm had been in discussions with Ericsson, Nokia and “new players” about vRAN and, like Orange, found that “from the point of view of the technology, it is good”. But he added: “In terms of operations there are some challenges. For example, if we use open hardware and software, who can guarantee the combination, as well as safety and configuration.” If any vulnerabilities were left to the operator’s staff to fix, “they cannot endure that kind of burden,” added Park Jin-hyo.
He added: “An operator could buy hardware from HPE and software from Ericsson, but if there are some issues, then who is in control? I think that in five years, there will be more virtualization in 5G compared with 4G, but it has to be based on guarantees around security and safety.”
Like many MNOs, SK Telecom is calling for greater cooperation with and between suppliers to address such issues in a cost-effective, unified way, rather than leaving every vRAN as an individual, hand-tooled project. Deutsche Telekom, at MWC, joined the chorus of operators criticizing large vendors for developing products that were specific to their own platforms. The need for the improved economics of an open, multivendor network has been a key driver for MNOs to support Facebook TIP – which includes a series of operator-backed incubators for start-ups with new approaches to the RAN. It will also be an impetus for cooperation in operator-dominated initiatives like ORAN Alliance.
However, CTOs like SK Telecom’s acknowledge that new architectures, while cheaper, may not be optimal for the demands of a mobile RAN. Even Intel has backed away from initial visions of running all RAN VNFs on standard x86 servers, and has invested in high performance coprocessors and FPGAs (field programmable gate arrays), which can be optimized for particular tasks. Notably, Intel acquired FPGA market leader Altera, while Qualcomm works closely with Xilinx in its own vRAN and server processor activities.
“The big difference with 5G is low latency, which requires a lot of processing power,” said Park Jin-hyo. “With open hardware based on FGPAs, I am not sure we can guarantee this kind of low latency and powerful processing. This means there are some limitations compared with the core network, where we can use common hardware from HP or Cisco. In the RAN, you need specialization for processing power.”
To get the operators excited about vRAN again, there need to be standards which ensure multivendor networks, but also far greater industry collaboration around ways to guarantee performance, security and deployability, even while reducing cost. These issues will be addressed as operators are pushed more urgently to adopt 5G, cloud and IoT services. All those will help make the case for vRAN, and will accelerate moves to make the vRAN architecture commercially and operationally, as well as technically, attractive. But the process will take at least five years, and it will result in a very different vRAN platform to the one that China Mobile outlined back in 2013.
Why has C-RAN lost its luster?
Initially, operators were excited at the prospect of slashing the cost of ownership of new networks by sharing resources flexibly among cell sites and replacing proprietary, specialized hardware with white boxes running virtual network functions (VNFs).
But of course, C-RAN proved far harder than most had hoped. Centralizing baseband functions and sharing them between a few radio units was fairly simple, an extension of existing base station hotels. But the dramatic gains would come from virtualizing those functions on commodity hardware, and sharing them with very large numbers of radio units.
That is difficult to achieve. White box servers need high performance processors or coprocessors to handle the demanding requirements of a RAN, so they are not as commoditized as many had expected. The lifecyle cost of RAN VNFs – inhouse or as-a-service – remains very unclear. Open source technologies are interesting, but to deploy them in carrier-grade environments requires a considerable amount of costly consulting or inhouse skills.
So the cost reductions are not as clear as had been supposed. Operators are increasingly looking for additional benefits beyond reduced capex and opex – an agile network which will make it easier to deploy new application, support new revenue streams, and move towards network slicing. However, the business case for many of those supposed new services, especially in the Internet of Things, is also uncertain.
And most importantly, there are two questions of balance, which will be critical to the success of a C-RAN in transforming an operator’s business, but will be hard to get right.
One is between physical and virtual functions – how much to support on the central, virtualized server, and how much to offload to the radio units as physical network functions. This is also important when considering how to keep legacy and virtualized RANs coexisting, probably for many years.
The other balance to strike – increasingly closely related to the first one for some MNOs – is between a centralized cloud approach and a distributed, edge-focused one. This has been primarily a decision related to cloud and IT infrastructure, but as operators look to converge connectivity and processing/storage resources, the two architectures need to be considered as one.
In a ‘classical’ C-RAN architecture, with hefty amounts of centralization and virtualization, there is maximum pooling of resources, which promises extreme efficiencies and agility. Everything can be shared on-demand between large numbers of sites over wide areas (hundreds of sites in China Mobile’s and Korea Telecom’s trials). This greatly reduces the need to over-provision sites, and the network can be scaled up and down by spinning up new virtual machines where they are required, according to changing traffic patterns and applications.
However, there are two barriers to this. One is the intense strain it puts on the fronthaul links between the baseband unit BBU) and the remote radio units (RRUs), if virtually all the processing has to be done centrally. Offloading some functions to the RRU can reduce the need to invest in massive fronthaul capacity and can reduce latency. This has led to the definition of eight different ways to split functions between BBU and RRU, within the 3GPP specifications, and the CPRI organization has defined its own set of five.
When the cloud and network architectures are planned in tandem, the decision on splits will be heavily related to an MNO’s views on its use cases. Despite the efficiencies of centralization, at least for operators which can access high quality fiber relatively cheaply for fronthaul, many aim to prioritize services that require low latency, context awareness and interactivity. Many of these relate to video content and to the Internet of Things, both seen as key drivers for new revenues in 5G. Such operators are more likely to aim for a decentralized, distributed approach to network and cloud resources?
For more information on the upcoming report on vRAN deployments and trends, from Rethink Technology Research’s RAN Service, please contact Simon Thompson on [email protected] or go to www.rethinkresearch.biz