The Open Networking Foundation’s (ONF’s) conferences always provide interesting foretastes of how the telecoms network of the future will look, and the recent Connect event in California was no exception. The group set out new platforms including an open source packet core and an evolution of its CORD (Central Office Re-architected as a Datacenter) technology. And operators from AT&T to Comcast shared the results of their efforts so far to drive virtualization based on open specifications.
AT&T, Deutsche Telekom and Telefónica – three leading lights in the push for open, virtualized platforms to transform the economics of telco networks – provided a joint update on what they call a ‘central office pod’ (CO pod). This is an extension of CORD – which is one of the ONF’s flagship projects to apply webscale approaches to networks – but uses a more modular approach which will be applicable to the more distributed, multi-tiered networks, complete with edge compute, which are envisaged for the 5G era. The pods are built on general purpose hardware and are coordinated using software-defined networking (SDN) and standard IT tools.
According to a study which consultancy Arthur D Little did for these three operators, a modular, virtualized, cloud-based architecture like this can save up to 40% in capex and 25% in opex.
Tom Anschutz, a distinguished member of technical staff at AT&T, told the conference about AT&T’s wish to drive cloud platforms into every part of its network over time, so that the same tools become the “bread and butter of services, communications or otherwise” and achieve a unified and scalable approach. AT&T has driven several open source initiatives itself, including the Open Network Automation Protocol (ONAP) and Open RAN (ORAN), and it also uses other open systems include some from ONF. These include the CO pod and the SND Enabled Broadband Access (SEBA), which the US carrier is trialling with 500 homes, providing low latency home broadband based on white box hardware.
Deutsche Telekom said its key goal in introducing virtualization and disaggregated networks is to reduce cost, particularly by increasing its supplier roster and therefore vendor competition. “The goal is redesign broadband access, focus on automation, bring in commodity and open hardware where it makes sense, and widen the spectrum of possible partners and vendors, but with a clear focus on delivering fixed and mobile access in Germany,” said Hans-Joerg Kolbe, a chief engineer. DT is also one of the operators – along with BT, Orange and SK Telecom – which is supporting the Telecom Infra Project’s open network efforts, incubating several disruptive start-up suppliers under the auspices of TIP.
Telefónica’s David Artuñedo Guillén – CEO of the OnLife Networks subsidiary – told the conference how the operator was combining virtualization and edge computing to enable new services and revenue streams. OnLife is specifically interested in producing and broadcasting TV content using 5G, and has trialled a system where producers sit in the central office rather than on-site at a news or sports event.
As well as SEBA and CORD, other ONF platforms include the SDN-based switching fabric, Trellis, designed for virtualized access and edge networks; the Open Network Operating System (ONOS) SDN controller; and the OpenFlow protocol, though this is being phased out in favor of the more functional P4.
Comcast was talking about its deployment of Trellis and ONOS, as part of its build-out of a Distributed Access Architecture (DAA). Comcast, along with AT&T, Deutsche Telekom, and Infosys, authored a Trellis reference design in April. The cableco says Trellis enables it to improve scalability and increase space and power efficiencies in its head-ends.
Also in the Santa Clara event, the ONF announced two new initiatives, the open sourcing of its Stratum switch operating system, and OMEC, the Open Mobile Evolved Core. Guru Parulkar, executive director of the ONF, said that evolved packet cores (EPCs) have become increasingly virtualized, and have moved to support control/user plane separation (CUPS). This allows key EPC functions, such as the S-gateway, P-gateway and mobility management entity (MME) to be disaggregated in software.
However, there have been two drawbacks to these early virtualized EPCs. One, they have used virtual machines (VMs), rather than the emerging technology of containers, which means they are cumbersome and inflexible compared to the upcoming new architectures. Instead of reworking existing physical functions as VMs, new cloud-native cores, which will be deployed for 5G from the early 2020s, will be created from scratch to run on cloud infrastructure, and will be able to break the core functionality down into large numbers of dynamically reconfigurable microservices.
Two, they have mainly been proprietary, and different virtual network functions from different vendors have not been able to interoperate in one core – or in the few cases where this has happened, such as NTT Docomo’s vEPC, it has required a large amount of expensive integration.
The aim of OMEC is to change this. Parulkar said: “That packet core has gotten more software-ized with VMs and put in data centers. That’s been a trend over the last five years. But it’s been all proprietary and closed source. We, with the help of Intel, have an open source version that’s CUPS compliant.”
He added: “If you already have an open source foundation, you can evolve it to become the next-gen for 5G. In the LTE world, proprietary things came seven years ago, and open source is coming now. But in 5G, open source and proprietary may come next to each other.”
OMEC is an evolution of the mobile strand of CORD (M-CORD). Intel contributed its inhouse EPC to the new project and is working with Sprint and T-Mobile Poland to start field trials of OMEC this year.
The ONF also announced that it would place the Stratum project into open source under the Apache 2.0 licence, establishing it as the foundation for a next generation SDN stack for operators. Stratum is a switch OS for SDN, which is neutral to the switch silicon and can use white box switching platforms.
“Ultimately, Stratum will become the substrate, the switching layer upon which all of ONF’s platform runs,” said Timon Sloane, ONF’s VP of marketing and ecosystem. “We announced it last year — an open source initiative to build a next-gen, SDN thin switch to run on hardware switches. And we’ve made significant progress.”
The biggest challenge, ONF speakers said, was to ensure Stratum can run on a variety of chip and white box options. Among those are the Tofino and Tofino 2 Ethernet switch-chips from Barefoot, which was acquired by Intel in June, and Broadcom’s Tomahawk. It can run on white box switches from Dell, Delta Networks, Edgecore Networks, Inventec, QCT and Stordis.
Because it avoids using proprietary silicon interfaces or closed application programming interfaces (APIs), Stratum should help operators to mix elements from different vendors and add new devices easily. The interface between the Stratum switch and its controller is based on the open P4 language so that the switch can be swapped for one from a different supplier, without the need for controller modification.
Operators working on Stratum, which was originally developed by Google, include NTT Group, China Unicom, Turk Telekom and The Netherlands’ KPN.
Amin Vahdat, VP and technical lead for networking at Google, said in a statement: “By making it possible to establish a ‘contract’ with the networking device, Stratum allows enterprises to unambiguously define system behavior and take advantage of the latest switching technologies without reworking the underlying control software.”
The Stratum project was kicked off in March 2018 based on the Google seed code. Other original contributors included switch-chip suppliers Barefoot, Broadcom, Cavium and Mellanox; plus China Unicom and Tencent of China and Dell EMC. Google said it would “help grease the market” by introducing the software into its production systems in 2018, even before the full code release. Stratum uses three open source protocols which Google helped develop — gNMI, OpenConfig and gNOI – as well as P4.
It has been hailed by Andre Fuetsche, AT&T’s CTO, as one of the elements that will enable the operator’s vision of a network based on white box switching, merchant silicon and P4. “This is more than just about lowering cost and achieving higher performance,” he has said in the past. “Frankly that’s table stakes. This is really about removing barriers, removing layers, removing all that internal proprietary API stack that we’ve lived with these legacy IT systems. Now we can bypass all of that and go straight to ONAP” – though critics of ONAP believe that will become over-complex, and redundant, in a cloud-native world where many of its functions will be driven down to the container level, leaving a far thinner layer where the OSS used to be.
The inclusion of Broadcom in the Stratum group was a feather in the cap for the ONF, since the chipmaker had previously been scornful of P4, and dismissed the idea that customers would want to program a network chip’s pipeline. If they did, it would offer its own C++ tools, it used to day. However, Broadcom is not expected to be running P4 programs natively on its chips, but to write translation layers. The same would go for network OEMs which still rely on ASICs rather than merchant silicon in their equipment. The biggest of those, Cisco, has not expressed interest in P4 to date.