Your browser is not supported. Please update it.

4 May 2021

Dish’s multiprocessor ambitions highlight Open RAN’s most critical challenge

By Wireless Watch Staff

It was another good week for the Open RAN movement, with several operators, notably AT&T and Telecom Italia, firming up on their commitments to deploying future 5G networks based on the open specifications. Perhaps even more important for the building of that key enabler of success, a broad ecosystem on top of a common platform, was Qualcomm’s announcement of a cooperation with Vodafone around Open RAN (see below).

As we have pointed out many times, for any tech platform to gain critical mass, but also to stay open to many vendors, it needs to have the support of major merchant chip providers – and those companies need to be able to match or outdo the performance of single-purpose ASIC chips, designed for individual manufacturers. Those dedicated ASICs can deliver highly optimized performance for demanding functions such as those of a 5G vRAN, but they are by their nature proprietary.

The trade-offs between common processors and dedicated, souped-up ones are at the heart of how most platforms have succeeded or failed in the past. In the modem itself, standards have been a success. Both IEEE for WiFi and 3GPP for cellular have set specifications that have been adopted by all, and have allowed multiple suppliers to innovate on the same foundations – though as Qualcomm’s huge market share in smartphone modems has shown, open platforms do not guarantee vendor diversity; that comes from multiple players having equal funds and engineering prowess to develop competitive offerings (which did happen in WiFi).

In the world of processors and accelerators for compute tasks, a common platform certainly exists, though it has historically been dominated, in PCs and servers, by Intel. But when compute tasks become extremely demanding, as in many AI applications, the commitment to the openness and scale weakens in the face of the need for differentiated performance. Hence the hyperscalers are not just trying to weaken their ties to Intel when they develop their own processors, such as Amazon’s Graviton – they are reverting to an old game of competing on the basis of underlying performance.

The mobile device world showed how an open processor architecture, on which multiple vendors could differentiate, was a possibility, with the ascendancy of ARM (even if very few suppliers outside Greater China managed to make a convincing business of competing with Qualcomm’s system-on-chip offerings).

On the infrastructure side, though, the field is still split between merchant processor and accelerator platforms from Intel, Marvell, Nvidia and others; the potential importance of the hyperscalers’ processors; and the ASICs that Ericsson and others claim are essential for optimal price/performance.

In an ideal world, of course, 5G Open RAN would run on a heterogeneous processor platform in which multiple vendors’ CPUs, GPUs, FPGAs and chiplets were all mixed and matched within a common fabric. This is a vision that some vendors are actively promoting, including, in its own particular way, Intel, which presumably believes it can shape the open cloud platform in the same way that Nokia hopes that, by embracing the disruptive potential of Open RAN to its own business model, it can lead and shape the outcome.

But on the processing side, the vision will be more challenging to achieve than a fully open vRAN stack in the layers above. Dish announced last week that it would run some of its Open RAN distributed unit functions on its existing Intel FlexRAN platform, some on AWS Graviton, and it followed up by saying it could also run some on Qualcomm processors in future. This was another indicator, like the Vodafone partnership, that Qualcomm is set to re-enter the infrastructure game in earnest and has the potential to steal the considerable thunder that Intel has made with FlexRAN.

However, Dish’s senior managers are inclined to make rather blithe statements about open networks. The idea of running the same functions across several completely different processor architectures in a seamless way is the stuff of heterogeneous Open RAN dreams, but it would currently add a big dose of complexity to Dish’s roll-out and to the challenges facing its suppliers. There would be plenty of work involved in adapting RAN virtual network functions from Dish partners Altiostar or Mavenir – which have been optimized, with considerable investment of time and expertise, for FlexRAN – for other architectures.

That is not to say this open portable environment is impossible to achieve, and it would certainly transform RAN economics if it did evolve with a wide variety of semiconductor players able to participate rather than one or two leaders. But this will take time, technical effort and the willingness of major market-makers such as Intel to be cooperative. In the meantime, it is important that Open RAN continues to score quicker wins – in encouraging an ecosystem of rApps and xApps for the RAN Intelligent Controller, for instance.

The bigger conversation about the open vision always brings back, for those of us who were following hi-tech industries in the 1980s and 1990s, the long journey to an open computer operating system. Time and again, vendors pledged to support Unix as this OS, only to develop semi-proprietary implementations that were tied to their own processor architectures. There were even feuds and breakaways among the open multivendor initiatives – the Open Software Foundation, set up in 1988 by several major computer suppliers, battled with the rival Unix International group for six years before they finally found common cause.

If that conjures associations with the various groups circling around Open RAN, the main lesson from history comes back to the processors. It was only when the Intel x86 platform became sufficiently high performance to power large computers cost-effectively that more open systems such as Linux were able to enable a multivendor environment (but one that was, for years, effectively tied to Intel, at least in the commercial world). Dish’s ambition to run its VNFs interchangeably across multiple processors highlights both the greatest disruptive opportunity of the Open RAN movement, and its biggest challenge.