Is “Open RAN” in 5G something we should welcome, or be afraid of? Is it all a part of the 5G hype, is it its own hype category, or is it not hype at all? What’s its relevance to telecom infrastructure overall? All of these are important questions, so let’s try to answer them.
“Open RAN” is generally applied to initiatives aimed at creating open-model 5G RAN technology, but you may hear more about the O-RAN Alliance, the dominant play in the space. The goal of the initiative is to create the broadest possible open specification for RAN elements so that a complete implementation can be defined. 3GPP RAN specs leave some components of the RAN (technically, 5G New Radio) opaque and Open RAN defines specifications and decomposition models for these. This permits best-of-breed element selection, and also broadens potential market entrants.
The latter is arguably the primary goal of Open RAN. Without it, the major mobile infrastructure vendors (Ericsson, Huawei, and Nokia) would likely lock up deals. That’s because those opaque components of the 3GPP spec would be implemented in different ways by smaller vendors and that would risk lock-in to a small and unfamiliar player. Needless to say, the major mobile infrastructure vendors are of two minds about this. On one hand, buyers fearing vendor lock-in could have their fears mitigated by Open RAN conformance. That could accelerate adoption. On the other hand, admitting others into the market isn’t exactly a smart competitive move for the three giants. A recent Light Reading story quoting Ericsson suggests that by the end of the decade, Open RAN could account for a fifth of the RAN market sales.
This answers our first question. If you’re one of the Big Three of RAN, Open RAN is a mixed blessing, but I think most would privately agree that they’re negative on the concept, forced to support it by competitive pressure. If you’re anyone else in either the operator or mobile infrastructure world, it’s a blessing. But what about its hype status?
5G RAN is already widely deploying, but if Ericsson is correct regarding the impact, then it will have a minimal impact on 5G in the near term. And is Ericsson correct? My own model suggests that the peak penetration of Open RAN depends on what you mean by it. Both Nokia and Ericsson have committed to Open RAN or convergence of their products with the spec, so if that’s counted as an Open RAN deployment, then I think we’ll see Open RAN hit 20% penetration some time in 2025.
But remember the idea was to create a truly best-of-breed model for RAN. If we assume that Open RAN penetration means the number of RAN implementations that actually take advantage of the Open RAN model to support multiple vendors, then I think it’s doubtful that we’ll ever hit that 20%. My model isn’t accurate as far out as 2030, but it seems to be plateauing at about 17% and wouldn’t even approach that until 2028.
But even if Open RAN has significant penetration, can we say it’s not hyped? As it happens, the issues with Open RAN hype may well be connected with the issue of 5G hype. It’s not that 5G is being exaggerated in deployment terms; it already dominates all the major markets. The problem is that 5G is usually characterized as a major new source of operator revenues, and since I doubt that it will be, the claim qualifies as hype in my book. I’ve blogged plenty on that, so feel free to look back if you want my reasons. So what would, could, should the role of Open RAN be in those new service revenue opportunities? That’s what decides whether Open RAN is hyped.
I’ve rejected the notion that just having 5G Core with network slicing was going to have significant impact on the mobile market. If nothing else can have such an impact, then you could argue that Open RAN isn’t very relevant except to mobile-infrastructure geeks. What would create a non-slice impact? Edge computing.
The big innovation in 5G, from an infrastructure and openness perspective, is the use of hosted elements rather than a fixed set of static appliances. Obviously hosted elements need hosts to run on, so presumably any 5G implementation would promote hosting. Open RAN defines more elements that rely on hosting, so it would promote more hosts. OK, that’s true, but there’s big “however”. Ericsson’s point in the article is that Open RAN is in fact effectively edge/cloud RAN, and that this model raises serious questions about the handling of the data plane.
Another 5G innovation is the separation of the control and “user” planes (CUPS). Functionally, the 5G UP is very much like an IP network, but it has a collateral role in CUPS, because there’s a link between the RAN implementation and the Core; think I-UPF (in MEC) and UPF (C-RAN) and also between the slice management (which is 5G specific) and the Core. Mobility management and slice management impact the UPF flow, which means that some UPF features would be “hosted”. That implies that the data plane of a mobile network would be hosted. You can implement a router as a hosted software instance, but it’s not likely to be the fastest and best option, which was Ericsson’s argument.
Solution-wise, the right answer would be to have a “router” that was a real, optimal, IP data device in all respects, but that would support some means of offering the things that mobile infrastructure needs, which is the GPRS tunneling and its control. Router plus 5G UPF-specific features equals 5G UPF.
The next-best approach would be to host the UPF features on an edge/core pool of servers with specialized chips to optimize packet movement. Intel’s x86 model is far from the only game in town even today; the article cites ARM, Marvell, Nvidia and Qualcomm as examples of other chips in use, and Broadcom offers its chips for white-box routers (DriveNets uses them) so they’d clearly be suitable. However, the use of a specialized resource pool could compromise the value of 5G as a means of driving early edge resource growth. Unless the edge applications needed the same special data-plane expediting or at least had another use for the special chips, the chip enhancements might make the edge resource pool too expensive for general use.
The solution to the general-edge-resource-pool problem is to use general x86 chips. As the article pointed out, Intel has taken the position that competition in the general-purpose computing chip space is high, and economies of scale in production are good. The former means that performance of these chips is likely to improve, and the latter that chip costs will be as low as they’re likely to be under any option. If we assumed that Open RAN penetration rates were modest over the decade, then we could assume that by the time there was a lot of interest in deploying hosted UPFs, the x86, AMD, and ARM options would likely work in most UPF missions.
You can see the problem here. Since Open RAN hasn’t taken off as fast as many had expected (or hoped), it’s not advanced edge computing much at all, and if data-plane performance encourages implementations that require special silicon, then it won’t promote a pool of general edge resources. Not only that, the requirement might well dilute the benefits of an open specification, since only a specialized implementation would be competitive with the Big Three vendors.
I’m a big supporter of open standards, and Open RAN, but I’m starting to wonder whether the market is outrunning the value proposition here. It may be necessary for the O-RAN Alliance to start looking at the specific question of suitable hardware for the UPF elements if we’re going to see Open RAN deliver what everyone hoped it would, which was an on-ramp to edge computing.