Your browser is not supported. Please update it.

1 June 2018

Challenges of virtualized mobile networks intensify at the edge

The complexity and management challenges of virtualized, cloud-based networks become even higher, and even less understood, when it comes to multi-access edge compute (MEC), or other approaches to integrating computing and storage capacity with connectivity at the edge of the network. This can support greater flexibility for operators to target resources where they are needed, and to support new services which require low latency response, context awareness or caching. Many believe they are naturally well placed, because of their real estate in central offices and cell sites, to provide the grid of locations to enable their own MEC platforms, and to support those of partners such as Amazon AWS.

But that will only intensify their NFV/SDN headaches in the short term. Cloud platforms like OpenStack, and virtualized network architectures like ETSI NFV, are struggling to adapt themselves to an edge-focused world, especially as the earlier discussions about telco virtualization were mainly focused on a centralized approach to maximize resource efficiency and reusability.

The OpenStack Foundation on the cloud infrastructure side, and the Open Networking Foundation (ONF) on the network side, are both trying hard to address the issue of the edge before telcos lose the momentum to other potential kings of the distributed cloud – from webscale firms building their own infrastructure, to vertical industries like utilities or manufacturers banding together to create platforms optimized for their specific needs and timescales.

At TM Forum and also at a workshop in Texas run by Facebook’s Open Compute Project (OCP), the ONF discussed the challenges of pushing its technology to the edge with only limited support from incumbent vendors. Timon Sloane, VP of marketing and ecosystem at the ONF, told the event that the major telecoms suppliers had not embraced the Foundation’s new edge-focused initiative, which was kicked off in December. This aims to define operator-driven reference designs to simplify and accelerate open source edge deployments. The ONF now sees a re-engineered supply chain, with new partners, as being important to achieving its edge aims.

“If you had looked at our partner list a few months ago, you would have seen all major incumbents on it,” he said. “Those incumbents have not stepped up, they have not shown us they are making that level of investment.” As a result, the ONF has shifted towards working with alternative partners, mentioning Radisys, Ciena and Intel.

When the Foundation made its new initiative public in March, its list of OEMs consisted of Fujitsu, Huawei, Samsung and one undisclosed company. At the workshop, Huawei was not mentioned. Some other large vendors are ONF members or collaborators but were never officially linked to the edge initiative. Those include Cisco, Ericsson, Juniper and Nokia.

Sloane believes operators are ready to start deploying edge platforms this year on disaggregated hardware and software, using open source technology and commodity servers. One operator is already moving production traffic onto an open source edge-based deployment of CORD, he said, though his optimistic view of timelines conflicts with many other signals coming from the OpenStack Summit, where telco edge compute was a major theme, and from individual suppliers and operators. As with NFV, there is caution about moving to large-scale production MEC projects until cost and standards questions are answered, and business cases refined. And with NFV, telco cloud and edge programs delivering the biggest benefits if they are implemented together, the delays in virtualization will have a knock-on effect on MEC too.

Sloane acknowledged that the trickle of operator MEC projects will not become a flood until there is greater choice and certainty in the ecosystem. “Operators are serious about deploying, this is not some lark. But the supply chain is not in place to deliver,” he said. “Operators have this supply chain problem today.”

This is what drove the ONF to kick off its reference design projects for access and edge, to accelerate uptake of CORD and other Foundation platforms. The references will include open source elements from ONF and also from other industry initiatives, including OCP and TIP – a sign that technologies driven by the webscale industry are converging with those defined by telcos. This may be threatening to the operators’ traditional competitive landscapes and ecosystems, but they have recognized that it is better to harness the power of Facebook and Amazon rather than fight against it. Now the vendors need to recognize the same thing.

Sloane said that, despite vendor reluctance, the edge reference designs are making good progress. The eight operators behind the initiative are developing “gold standards or blueprints for how you put these pieces together and then the operators are going to procure or buy off these reference designs”, providing vendors with product roadmaps.

“Operators are committed to reconstituting a supply chain that is aligned with vision of white box, open source and disaggregation,” he added. “They want a supply chain that is excited about this possibility, so don’t come back with completely proprietary solutions or open source buried by proprietary APIs [application programming interfaces].”

But even when new vendors are fully on-stream, and old ones adjust their thinking, Sloane is still echoing the caution of the operators at the TM Forum event on cost reduction potential. Telcos are “not expecting to spend less money”, he insisted, but will spend a similar amount with a different mix of hardware and software, and with improved return on investment thanks to greater business flexibility.

This will be unwelcome news to operators like Telus with aggressive cost reduction targets, though Sloane’s comments do chime with those of some operators – like Orange – which have moved on from seeing virtualization mainly as a cost cutter, and will now justify it primarily in terms of new revenues. However, it takes longer to develop and launch new services and applications platforms, especially if these target new markets such as Industrial IoT, than it does to reduce costs, and the results are less easily measurable. So the shift from cost to revenue is likely to delay NFV and edge projects further, while operators work out their business cases – and while they wait for the next great hope for slashing costs, automation.

The intense focus on edge compute was also reflected at the OpenStack Summit, where there was a dedicated stream for edge issues. It is strategically important for the OpenStack platform to make itself indispensable for telcos’ cloud deployments right to the edge, since so much momentum in the more general cloud computing world is shifting to containers and Kubernetes (OpenStack works with these communities, but does not direct their agenda, and in the area of containers, has launched its own project, Kata Containers, its first not to be based directly on OpenStack).

But there was discussion at the Summit about whether telecoms is actually the only vertical which remains fully focused on OpenStack, and it is certainly the most important one for the platform – hence the criticality of extending its influence to the edge.

The telecoms vertical may be the biggest deployer of OpenStack cloud infrastructure, but operators still have their reservations about the platform, with some complaining that it takes too much time and money to implement the technology to a carrier-grade standard, and that the huge variety of community projects now running under the OpenStack umbrella risks fragmentation and makes it hard to decide where to focus resources.

Some operators still believe OpenStack – or perhaps any open source system – cannot meet the rigorous demands of telco networks and are cautious about taking such platforms outside the data center. Thus there was a strong sense in Vancouver of the OpenStack community seeking to convince telcos that it has addressed their concerns with regards to security, robust performance, mature tools, integration with other important technology efforts from ETSI to Kubernetes, and of course the mobile and multi-access edge.

The edge cloud opens up a new opportunity for OpenStack in telecoms, where operators are increasingly using the technology for NFVi (infrastructure), and are also increasingly interested in integrating their connectivity with edge compute capabilities to enable new services. But there are doubts about the suitability of the platform in the smaller footprints of edge servers, where OpenStack may prove too unwieldy and resource-hungry to deploy.

Last autumn, the OpenStack Foundation’s executive director, Jonathan Bryce, said edge computing had emerged as a “surprise” use case for the open source platform. Enterprises – particularly telcos – which are using OpenStack as the basis of their private clouds are starting to consider how to distribute those cloud platforms to support services close to the user.

In the first quarter of 2018, the Foundation’s Edge Computing Group released a white paper, Cloud Edge Computing: Beyond the Data Center, which looks at the specific requirements of edge computing and its requirements for OpenStack.

There is also a clear focus on edge in the latest release of OpenStack, called Queens.
Queens included the work of several edge-focused projects, such as OpenStack LOCI (Lightweight Open Container Initiative) and OpenStack-Helm. Both these simplify the deployment of OpenStack services at the edge. The former provides LOCI-compatible images of OpenStack services, allowing them to be deployed by a container orchestration tool such as Kubernetes.

The latter is the result of a project led by AT&T, SK Telecom and SAP. It uses the Helm packaging system, which is a way to map a complex Kubernetes application, and define the interaction of containers. That enables a small footprint OpenStack deployment, with Kubernetes controlling it in an automated way. The project had its first full release in February and AT&T has been trialing it.

Now the Foundation needs to convince a wider group of telcos that it has the right platform to support their significant and complex edge cloud deployments. The US is leading the way, as in many areas of virtualization and cloud. Bryce said: “What AT&T and Verizon are working on with OpenStack is all of the computing and networking infrastructure that sits between a smartphone or smart car or some kind of IoT device in a house and whatever data center that data is going back to. It’s a very big opportunity and footprint these guys are working on. It has different characteristics, and these are some of the things we’re a little bit behind on the OpenStack technology front.”

Partnerships will be important to catch up. “When you’re dealing with a centralized data center, it can make sense to create or use proprietary solutions and utilize public clouds, but when you’re dealing with an edge computing scenario, leveraging an open infrastructure makes so much more sense,” said Ildiko Vancsa, ecosystem technical lead at the OpenStack Foundation. “Many edge workloads today run on Linux or in virtual machines.They may evolve for simplified lifecycle management, or they may be superseded by a next generation of applications. No one technology will emerge alone to provide a complete solution for the edge.”

Some of the groundwork has been laid, especially in relation to ETSI MEC. Both ETSI and OpenStack say they are keen to avoid a stand-off – as seen in the area of MANO – in edge compute. ETSI’s MEC group says there is no official agreement with OpenStack because they are working on complementary, not overlapping, issues. And some open source players have worked hard to cooperate with MEC.

For instance, Red Hat joined the ETSI MEC group in 2016 to help define use cases and architectures and ensure these remained complementary with open cloud platforms. There will also be an interest in trying to limit the influence of Amazon AWS, which has its own close alliance with MEC focused on its Greengrass edge-based IoT platform.

There are other open source edge movements to consider too, and some may be less friendly than ETSI. The Linux Foundation is hosting two projects – the ONF’s CORD and the AT&T-initiated Akraino. CORD claims it is now the “de facto choice for edge compute” among operators, but says it is not working with ETSI because, as the ONF’s Sloane put it: “CORD gives you all that natively; you’re going to get the equivalent of edge computing. We believe open source is moving at a pace unlike anything seen. Open source is having a dominant impact.”

Akraino is closely tied to OpenStack but not to ETSI. It aims to define an open source software stack to support carrier-class, high availability cloud services optimized for edge computing. Its first code will be released in the second quarter this year. “Akraino Edge Stack, coupled with ONAP and OpenStack, will help to accelerate progress towards development of next generation, network-based edge services, fueling a new ecosystem of applications for 5G and IoT,” said Mazin Gilbert, VP of advanced technology at AT&T Labs, which contributed some of the code.

OpenStack does not have to worry about the edge compute architecture which prevails with operators, since it is a strong candidate to provide the infrastructure for any of them – though too much fragmentation could slow the whole market down. But it does have to build operator confidence in its capabilities, compared with inhouse infrastructure, or AWS and Azure public clouds.

Bryce said in a recent interview that the Foundation is now working hard to tailor OpenStack work to specific use cases, one of the areas being edge computing and NFV.
These use cases have generic needs for automated compute, storage and networking, but there are differences of implementation. For instance, edge compute needs self-service capabilities, automation and APIs, and has to take into account large numbers of server deployments, some of which may be in difficult environments.

“One of the big issues at the edge is that you don’t have one environment, you have a lot of different environments, so that automation, that zero-touch management is very critical so this is being built directly off of AT&T and SK Telecom requirements and both of them are trialing it,” Bryce said.