Carriers Push Datacenter-Style Virtualization

Move is viewed as way of adding flexibility and better security, and of tapping into IoT growth.


The world’s largest telco carriers are leading a broad movement to bring data center-style virtualization to the core of their telecommunications networks. In an industry known for being extremely conservative when it comes to change, this one appears to be significant.

The move has set off a scramble among a number of companies for unified control and forwarding plane designs, starting from the low to midrange systems and spanning everything from the access and aggregation layers of the fixed network.

Beneath the hood of this shift is an effort to tap into the growth of the IoT and the innovation surrounding it.

“The carriers are looking at these IoT-like services, and at the same time looking at how and when they transition off of their 2G networks, and virtualization is an absolutely key technology,” said Ross Cassan, director of product marketing for mobility infrastructure at Spirent Networks. “We are seeing all of our carrier customers really come to grips with that.”

Spirent’s platform emulates a Narrowband-IoT network from end to end, verifying whether the network can handle the expected IoT traffic and, if so, what kind of service-level agreements (SLAs) can be guaranteed and relied upon. The company worked with the Brocade software group formerly known as Connectum on that.

“When you do that, you can see where you might simplify the protocols, and simplify the silicon at either end—the end device or the core of the network,” said Cassan.

Using virtualization, carriers can create a ‘network slice’ for IoT for a given set of customers and applications, like utilities and power meters, or banks for transaction processing. It’s this work that actually makes the IoT business model viable for carriers.

Cloud seeding
The cloud-based software revolution, together with the processing power gains the semiconductor industry has made, have enabled this “central office re-architected as a data center.” Known as the Mobile Central Office Re-architected as a Data Center, or M-CORD, the name is consistent with the alphabet soup of acronyms that has always dominated this market. M-CORD is just one of a multitude of efforts created by the NFV (network function virtualization) Congress to iron out how equipment makers, standards bodies, and carriers will get this very seismic shift in telco network architecture accomplished.

“All of the things that people worked on for the last 7 to 10 years are the same general techniques” the carriers can use to scale at 5G, said Raj Singh, vice president and general manager of the network and communication group at Cavium. “The carriers are saying, ‘Hold on a minute, if I can make this work it makes my life really easy. I don’t have to make custom hardware…By definition, virtualized functions are scalable.”

ARM and Intel enabled this when their cores gained the ability to support hypervisors, Singh said. “You needed to be able to run hypervisors and containers and dockers.”

That happened in the new cloud-based software data centers. Cavium makes an ARM-based network server chip solution that work in this mode. The Spirent/Brocade solution is Intel-based. Both are enabled by extensions that allow the system to run the kernel-based virtual machine (KVM).

ARM has been building out its IP portfolio for the customers in this ecosystem for the past few years, having recognized that the carrier networks were poised for change. The company coined the “Intelligent Flexible Cloud” for “intelligence across the network, with different combinations of storage, acceleration and compute, rather than in previous generations having intelligence located at either end of a rather dumb network,” according to Bob Monkman, a senior segment marketing manager with ARM.

Cavium’s Singh points out that the semiconductor industry crossed a threshold in terms of compute power and latency, which makes virtualization and intelligence across the network possible. “After about 28nm, the silicon was capable and quick enough to perform this, to allow for the extra level of abstraction that is virtualization in the network,” he said. “At 28nm and after that, these network ICs were up to the task. It was always feasible. But it became possible.”

Singh notes that as carrier networks improved from 2G to 3G and 4G (LTE), they added capacity and feeds. Hardware scaled in the form of fixed function appliances, increasing cost and complexity all along.

“Why do we need to change the network architecture now? Why can’t we just continue along the same path?” Singh asks. The answer is the gigabit-per-second capacity required for 5G. “That’s 20 times the capacity per sector—100 Gpbs, in each direction, and not by two. It’s by eight. That is impossible, because it means 20 cables going to each one of those radio heads. It’s clearly not a feasible way to go building networks in the future. Suddenly that becomes very important, if you can virtualize the network for the radio access.”

Disaggregating the radio access network (RAN) allows carriers to create a new front-haul network with a lower bandwidth requirement.

“In the network topology of the future, the core of the network is in virtualized baseband networks,” said Singh. Cavium and ARM together have become expert in silicon that can run multiple virtual machines, each of which is separate from others and secure. That has always been one of the advantages of virtualization technology, and it is in widespread use throughout datacenters partially for that reason.

The cellular side
What happens on the server is only one piece of the puzzle. The cellular infrastructure is undergoing change, as well.

“Cellular infrastructure is arguably one of the most dynamic segments right now,” says ARM’s Monkman. “Certain equipment types will be driven much more towards latency and throughput with stringent power budgets. These cannot be met with generic, virtualized compute only. You need a mix of the right core, the right interconnect [depending on whether the ASIC is for the Aatenna, basestation (BTS) or mobile edge computing (MEC) boxes]. As functions disaggregate from the BTS, real time offload and feature rich APIs are key.”

Some ARM SoC vendors are delivering more than 100 cores in a given device for this new style of network compute power.

“ARM SoC vendors deliver a wide range,” Monkman said. “That means 2 cores in some applications, all the way to 100+ cores. It really depends on the use case, some being remote radio heads or low end routers, through RAN and even EPC proof points. Our licensees deploy many-core SoCs, with virtualization hooks, different combinations of compute, storage and value-add accelerators for workload optimized solutions. The ARM business model is well suited to the range of use cases from RRH, antennae, IoT gateways, storage controllers, up to high-performance servers. A high percentage of cellular equipment edge network based equipment is, in fact, ARM-based today. But this is new equipment that has transitioned over from MIPS or POWER-based designs.”

And this is just the beginning of this transition.

“For MEC, and for the core network, there is certainly a desire to use more generic compute requirements, reuse of software over platforms, faster deployment and development cycles, and CoRD is one of the frameworks that is set up to enable this type of agile development and deployment,” said Monkman.

He cites ARM’s activity in Linaro, the open-source software community, and more specifically helping to define the right software interfaces to allow network offload and ensuring NFV compliance for these containerized/virtualized environments. In OPNFV, ARM has been delivering an ARM-based NFVI platform for three releases now, through the Armband subproject, he noted.

The “M-CORD” project, meanwhile, began at Stanford rather than in the labs at Verizon or AT&T.

Not so fast
But carriers around the world have a large 2G infrastructure today. They only want to decommission it as necessary, say numerous industry sources.

“Virtualization allows them to start bringing over customers and offer them IoT services on their network, and then scale it as the financial and business models evolve and mature,” Cassan said.

China Mobile has an Open NFV test lab that Cavium and ARM are supporting, as well as the Swedish network software stalwart Enea. Singh said Comcast, Google, SK Telecom and China Unicom, the mainland’s other giant carrier, are embracing and supporting M-CORD too.

“One key emerging area is for containerized microservices architectures—meeting real-time, latency-sensitive traffic needs, and re-defining the hierarchy for the infrastructure, is 5G slicing,” said ARM’s Monkman. “ARM views this particular emerging segment as a blend of the edge access requirements and the core network, and ARM is focusing on where NFV equivalence across architectures is key—while balancing the performance with the correct level of offload for edge applications to meet the 1ms turn-around targets for mmWave applications such as virtual reality, augmented reality and artificial intelligence.”

After watching other technology segments take advantage of gains in performance and power and increased connectivity, driven in part by the IoT, the large telecommunications companies are finally poised for change. How quickly that plays out, and in what markets or regions, remains to be seen. But for the first time in years, big changes are at least being discussed and tried out. And that, in itself, is worth watching.

Related Stories
The Case For Narrowband-IoT
IP, software for existing LTE networks, in sub-$5 modules.
How Will 5G Work?
RF chip giant Qorvo explains how 5G will get deployed.
Will Hypervisors Protect Us?
They may not be a silver bullet, but they are a good first step when it comes to securing cars and the Internet of Things. Problems start when people believe the job is complete.
Virtualization Revisited (May 2016)
As Moore’s Law becomes more difficult to follow, virtualization architectures are getting a second look.