Open P4 helps to make white boxes truly carrier-grade

Every week brings a new development in the rapidly moving area of open source platforms for telcos, many of them driven by operators which want to free themselves from the stranglehold of their established vendors and shake up their cost base. In this context, one of the demonstrations at this month’s Mobile World Congress will be particularly important. Run by the Open Networking Foundation (ONF), it will show the latest code written for the P4 programming language, running on powerful white box switches based on merchant switch-chips from Barefoot Networks, Cavium and Mellanox.

The operators supporting this demonstration include AT&T, China Unicom, Comcast, Deutsche Telekom, NTT Group and Turk Telecom, and significantly, they are joined by Google. Significantly because, while most of the telcos are looking to open source to slash their costs and defend their market position with more flexible resource usage, the open technologies can also be seized by Internet or other players (including cablecos like Comcast), to disrupt the established order.

Operators know that, however eager they are to support platforms that enable a more innovative ecosystem and lower barriers for new companies to enter their supply chains, they need to be agile in embracing those platforms before a new challenger does so. P4 is an important element of the emerging open telco world because it seeks to address some of the telecoms industry’s concerns that OpenFlow was not sufficiently carrier-grade to support the most demanding network functions.

The lack of confidence that open source platforms with enterprise roots, also including OpenStack, can meet telco requirements, is a significant barrier to change, but one that is being broken down steadily. A few years ago, most telcos were interested in OpenStack and its derivatives, but clung to more traditional standards bodies like ETSI for their core technologies. As more telecoms players have contributed to open initiatives, the balance has started to change – while ETSI NFV is the leading open foundation of carrier virtualization, at the higher layers, including management and orchestration, OpenStack-based options like ONAP are gaining ground on ETSI’s own OSM.

In the area of white boxes – with which operators aim to replace expensive proprietary hardware as part of their move to NFV and software-defined networks – there has also been significant effort to make the platforms more robust. The ONF, which is also the home for the CORD project, previously placed the OpenFlow protocol at the heart of its work, but has now also adopted P4, which was pioneered by Barefoot, to make up for some limitations in OpenFlow.

The ONF and its supporting carriers now say they will use their MWC demonstrations to raise awareness and support for P4 and ensure a broad ecosystem develops. One of the most important companies to attract would be Broadcom, the largest switch-chip maker, which is said to be showing interest in P4 as vendors move away from expensive, specialized but inflexible ASICs and look for merchant silicon which is programmable, while also supporting extremely high performance levels.

Broadcom took a step towards openness last month when it announced an open source Software Development Kit (SDK) for its Ethernet switch-chips, and open sourced the table APIs for its switch-chips, opening up its forwarding plane, though not making it fully reconfigurable.

The glaring exception to this trend is Cisco, which is still firmly wedded to ASIC-based systems and showing no interest in P4. But it may find its position under threat if the ONF achieves its aim of building a broad community of open source white box OEMs, and these secure high level contracts with large operators. Chip and vendor partners involved in the MWC demonstration are ARM and Intel, plus Ciena, Netsia, Radisys and Mariner Partners.

Quanta and Delta are among the existing white box vendors, but other, bigger names are needed to convince telcos that a robust ecosystem will develop – and there need to be far better proof points in mobile.

So far, ONF has P4 code for access networks in field trials but says software for the mobile core is one or two years behind that stage. The end goal is that P4 should help enable a fully open network, based on commodity hardware, from central core to distributed edge cloud.

In the edge cloud, the ONF aims to make its key projects – CORD (Central Office Re-architected as a Datacenter) and ONOS (Open Network Operating System, an SDN controller – the central enabling technologies, rather than efforts like ETSI Multi-access Edge Computing (MEC), which have come out of traditional standards processes. Not that the organization wants to set up a damaging conflict with the old ways – one of the features of its MWC demo will be interoperability with ETSI MEC, “demonstrating the flexibility of the CORD architecture and the effort to work with the broadest possible community of stakeholders”, it said.

At MWC, the showcase will be based on a CORD POD, an ONOS-controlled P4 switching fabric, and a range of operator-defined use cases.

The Foundation claims that CORD is “defining the edge cloud of the future” with a range of enhancements. These include:
• New 100x performance, cost, power and space savings enabled by hardware-based VNF (virtual network function) acceleration with SDN
• Per-subscriber real-time orchestration, with XOS chaining together VNFs on servers and programmable P4 switches to instantiate custom services for individual subscribers
• Hierarchical orchestration and automation, with ONAP as a global orchestrator and CORD as the intelligent edge cloud infrastructure with real-time subscriber orchestration
• The first commercial grade virtual function (vMRF) with complete lifecycle management and policy-driven automation via ONAP on CORD platform
• Mobile Access Edge Compute services including VoLTE, edge video processing, and advanced VR/AR requiring high bandwidth low latency services on CORD platform.

And to highlight perhaps the biggest attraction of open source for telcos, it added that “17 collaborating organizations created a single unified 5G demo in just three months”.

The CORD-based solution will demonstrate several other benefits operators are seeking, including multivendor switch networks with full programmability, and the ability to support network slicing (this will be done via integration with ONAP and Netsia Programmable RAN. ARM will contribute a facial recognition use case based on GPU acceleration, while Barefoot and ONOS will demo VNF offloading, with the 3GPP user plane running in P4 hardware. Radisys’ use case will show local break-out, VoLTE and virtual reality applications on a CORD-based MEC platform with ONAP-based lifecycle management.

“We’re at an interesting inflection point,” Timon Sloane, VP of marketing and ecosystems for ONF, told EETimes, hailing the MWC demo as the kick-off for “a second generation of software-defined networks”. “We learned a ton from OpenFlow, but it has limitations, so the community strategically shifted to P4 and the P4 runtime to solve problems in a more comprehensive way.”

He acknowledged that ONF is no longer actively developing OpenFlow, which has been used by Google and telcos to access the data forwarding pipeline of network ASICs – but cannot access all the functions, and does not allow the pipeline to be programmed, unlike P4. Critically, OpenFlow did not support full multivendor interoperability because adjustments were needed for each ASIC.

One of the first operators to support P4 was AT&T, which has worked with Barefoot as well as Broadcom on white box switches. Late last year, it also introduced its dNOS (Disaggregated Network Operating System), an operating system for white boxes, which it wanted to open source. It has now announced that dNOS will be hosted by the Linux Foundation, along with ONAP and another of its projects, XRAN – though it is contributing this system at a far earlier stage and expects the bulk of the code to be written by third parties. ONAP is based on a major AT&T inhouse project in management and orchestration, called ECOMP (which was then merged with some work by China Mobile called OPEN-O). About 8m lines of code were contributed.

But dNOS is still at the early stages of existence. Chris Rice, SVP of Domain 2.0 Architecture and Design at AT&T, told LightReading: “There’s a fair amount of code that we have and that we got when we acquired Vyatta. What we have to figure out is what is the right architecture and what pieces of that code make sense to donate.”

So far, dNOS is not part of the Linux Foundation’s new LF Networking organization, which it announced last week to improve coordination between its various telco-related projects, but Rice is clearly a believer that it should be included in that umbrella framework (though every project gets to decide that individually).

AT&T said that the dNOS project will support existing network protocols, as well as newer tools like P4.

“Our goal with open sourcing the dNOS project is to create a community around an open framework to software-enable industry standard white box hardware designs, such as those contributed to the Open Compute Project (OCP),” said John Medamana, VP of packet optical network at AT&T. “We’re excited to work with The Linux Foundation to bring this concept to reality. We invite others to join us to build the community and support this effort.”

“There is no doubt about the eventual direction for the industry and the emergence of white box as a major design element in new SDN networks for service providers,” wrote Medamana and Rice on a company blog last year. “It signals a new direction and way service providers will design their networks going forward. Just as in the webscale industry, initiated by the need for open interfaces to collect data that drives automation, white box and its new ecosystem are finding a home in the telecom industry. AT&T is looking at options in the industry where community-based work in this area could take place.”