Your browser is not supported. Please update it.

29 July 2019

AT&T makes “public cloud-first” pledge for IT, networks sure to follow

The move to the cloud is still years away for some MNOs, but those in markets which are already embroiled in 5G price wars, from the UK to South Korea, are likely to feel the pressure to make the move to a cloud-native core in the next two or three years. The UK’s Three and BT/EE have already outlined their plans in some detail, as have AT&T and Verizon in the USA, Rakuten in Japan and others.

One of their key decisions will be whether to build their own cloud infrastructure to host the core and RAN, or rely on a public or hybrid cloud from a partner. AT&T has set an important precedent by talking up a “public cloud-first” strategy, which already involves various deals with Microsoft, IBM and Amazon AWS. Three is using a mixture of public and private cloud, claiming to be one of Azure’s largest UK customers thanks to its deals with the Microsoft unit.

This contrasts with the decisions of some other operators which are advanced in the cloud. Even Rakuten, the Japanese MVNO-turned-MNO, which is deploying a fully cloud-native 4G/5G network from scratch, is building its own cloud. The ecommerce and content company runs most of its current services on a partner’s cloud, but for the mobile network, which makes particularly taxing demands on the cloud infrastructure, it is building a specialized platform with the help of Red Hat (software), HPE (hardware) and Cisco (integration). It has a group of RAN and core vendors whose virtual network functions (VNFs) will run on top of that cloud, but acknowledges that it is currently a lengthy and complex process to tune the NFV infrastructure and the VNFs to achieve performance – especially in the RAN – that is equivalent to that of conventional, dedicated appliances such as physical base stations.

This will present a major challenge for public cloud suppliers which want to expand their business into hosting telcos’ networks, since the goal of cloud-native, containerized deployments is not to have to spend months on tuning and calibrating, but for the VNFs to run almost instantaneously, in a common manner, like applications on a phone or PC.

AT&T will no doubt be discussing such issues with its cloud partners, Microsoft Azure, IBM and Amazon AWS. The US telco has had a close partnership with AWS for several years, though this mainly relates to joint offerings for enterprise customers, in which the two companies provide a combined solution. This was significant when AT&T announced the deal because it showed the operator – like many of its peers – backing away from the idea that it would operate its own cloud to support enterprise or consumer services. Instead, it will focus on what it does best – deploying and running networks, implementing security and device management and so on – while calling on AWS for the cloud capabilities, rather than competing with it.

The same debate now seems to have been extended to where AT&T will host its additional enterprise offerings as well as its own internal services, including its BSS/OSS, and eventually its core and virtualized RAN. Last week, the company said its target was now to become a “public cloud-first company by 2024”, as John Donovan, CEO of AT&T Communications, put it.

The shift to the public cloud is a logical culmination of a process which began last year when AT&T sold its own cloud infrastructure for $1.1bn, while at the same time increasing its reliance on cloud services.

In a blog post about AT&T Commnunications’ deal with Microsoft, he wrote: “It’s no secret the cloud, AI and 5G will drive the next surge in innovation … For that, there’s no stronger technology pairing than the capabilities of Microsoft’s cloud and AT&T’s network. That’s the significance of our new extensive, multi-year alliance.”

So far, though, Microsoft is only the preferred cloud provider to support AT&T’s non-network applications, and Donovan has excluded network functions from the 2024 deadline, while making it clear, in other comments, that the ultimate goal is to move cloud-based network elements to third party infrastructure too over time, even if they are initially deployed on AT&T’s own infrastructure.

The two-step process would make sense, giving the telco full control in the first phases of deploying a cloud-native core and RAN, when it is almost certain – like Rakuten – to have to do extensive tuning and optimization. Over time, as these network functions mature and are better understood, the end goal of a fully automated, programmable, dynamic network should be achieved, and at that point, there is no reason not to move the RAN and core to the public cloud too. That, however, could be very late in the next decade, even for a technologically advanced operator like AT&T.

Indeed, it may find itself leapfrogged, along with other cloud pioneers like Telefónica, by a breed of MNOs which move more slowly towards 5G and cloud networks, because they do not need them yet and/or lack the technical capabilities, funds and confidence. This second wave of 5G providers, some in emerging economies, may make their move once others have taken on the hard work and risk of making network functions fully cloud-capable and automated. At that point, cloud economics will come into play with strong results for MNOs with very tight budgets, such as those in India.

But for the next few years, the cloud economics that MNOs dream of – delivered by automation, commoditized hardware and simple containerized software – will seem very elusive, and the near term financial improvements will come from moving IT and customer-facing applications to the public cloud, rather than the network itself.

So AT&T says it will consolidate its data center infrastructure and operations in order to move its workloads to Azure, reducing costs and deploying new services more quickly, particularly in leading edge applications such as machine learning and advanced security.

On Microsoft’s side, it wins a significant new customer (the deal is reportedly worth $2bn), plus an endorsement of the viability of the public cloud for telco operations. It will also gain access to AT&T’s 5G capabilities. Donovan said: “Our 5G capabilities are already spurring innovation in manufacturing, healthcare and entertainment. Now, Microsoft will also tap into the innovation and resources our 5G network delivers. That includes collaborating to design, test and build edge computing capabilities. Edge computing is the future – and together we’ll be at the forefront.” The two companies are planning to launch joint services in areas like Internet of Things (IoT) and cybersecurity – applications for which, until last week, AWS might have expected to be the leading cloud partner.

The AT&T Business division announced its own cloud alliance and will work with IBM Cloud, moving some of its business applications to that platform while becoming IBM’s primary provider of software-defined networks (SDN) in return. AT&T Business will also use the open source Red Hat software (now owned by IBM) to manage workloads for its enterprise customers and will collaborate with IBM on multicloud capabilities for 5G, edge computing and the IoT.

For now, the areas of AT&T’s network which have already been virtualized are running on its Network Cloud (formerly called AT&T Integrated Cloud or AIC), which has been migrating to a more cloud-native approach with the addition of Kubernetes containers partly enabled by an alliance with Mirantis. That three-year deal was announced in February and will help AT&T build out the next  generation of its Network Cloud, this one focused on 5G.

AT&T has been building this Network Cloud using software from an open source project it helped to initiate within the OpenStack Foundation, called Airship. That was kicked off last year with SK Telecom and Intel and aimed to make it easier to build and manage a cloud. AT&T is using the software to make it practicable to roll out a large number of data centers and manage them on a single lifecycle – the Network Cloud has more than 100 data centers so far.

Now it plans to refresh the cloud infrastructure to align Network Cloud better with its impending 5G upgrade, and it has hired Mirantis to help. The suppliers CMO and co-founder, Boris Renski, said: “Importantly, what drove this rearchitecture and name change is the necessity to also roll out 5G workloads.”

The Network Cloud was built using OpenStack cloud infrastructure software. Mirantis has a commercial OpenStack distribution and is an active member of the Foundation, and has also worked with AT&T over several years, to it was perhaps a logical choice to help with the transition to a next generation Network Cloud. The big difference will be that Network Cloud for 5G will be built on OpenStack using containers, managed by the Kubernetes container management systems, not virtual machines (VMs) as before.

Of course, many workloads will still be running on VMs in the legacy platform for years to come, and the first container-based workloads will be specifically related to 5G.  But AT&T’s pace of change is rapid. According to its associate VP for network cloud software engineering, Ryan Van Wyk, the telco is deploying OpenStack on Kubernetes in more than 20 regions to date.

The platform is also supporting AT&T’s FirstNet public safety network, which is running as a workload on Network Cloud, demonstrating security advances made by the Kubernetes community. He told LightReading: “We’re progressive, we’re on the bleeding edge. The 5G core and architecture we’re implementing – we’re doing it for the first time in the world. When you’re pushing the capabilities of the available software and you’re in the front end of that, you need to innovate fast. We believe the communities around open source projects are the way to do that.”

Among the 5G-oriented virtual network functions (VNFs) to be deployed on the AT&T cloud are the virtualized evolved packet core (vEPC), RAN backhaul, traffic shaping services, customer usage tracking, smart voicemail, video streaming and many consumer facing services. However, even AT&T is currently cautious about putting a timescale on virtualizing its RAN, at any scale, in the cloud.

Mirantis CEO and co-founder Adrian Ionel explained that the company’s platform allows Kubernetes to be run on-premises, on bare metal, or in the cloud. For AT&T, that Kubernetes base will support OpenStack as a workload on top of the container orchestrator. OpenStack is needed to support NFV and to orchestrate VNFs from different vendors on one cloud. Mirantis expects the platform to run a few thousand nodes this year, and then scale to 10,000 nodes over the next three years, and more than 20,000 nodes “in the years to come”.

Mirantis’ work also involves integration of code from its Mirantis Cloud Platform (MCP), which supports VMs using OpenStack, containers using Kubernetes, and bare metal, all on the same cloud. Last year, it added its MCP Edge platform. “This is really about Kubernetes taking a prime role in the future infrastructure of a gigantic carrier,” Ionel said. “The scale of this is really staggering.”

Mirantis is also working with India’s Reliance Jio to run OpenStack on top of Kubernetes, but companies can deploy different combinations too – for carmaker Volkswagen, it is running Kubernetes on OpenStack in an on-premises environment.