An Interview with Francisco “Paco” Martin Pignatelli, Vodafone Head of Open RAN and Cristina Rodriguez, Intel VP and GM, Wireless Access Networking Division
December 2022
Joe Madden, Mobile Experts
JOE MADDEN: Paco and Cristina, thank you for joining me to talk about your progress in the UK and your plans for Open Virtual RAN. As I publish market research in this area, people are constantly asking me about performance and power efficiency. Let’s start by talking about the performance of your sites in UK and your Golden Cluster testing. How well is it working, compared with your traditional RAN Networks?
PACO MARTIN: Thank you Joe. We started our journey in Open RAN a few years ago, and it’s been progressing quite well since then. Obviously, there were challenges we could not even anticipate four or five years ago, but the reality is that today the situation is much better.
Our main project is now in Vodafone UK as you said. We are planning to deploy 2500 sites and we are well underway on this project. We started with the first site close to a year ago, which wasn’t the final configuration, but we decided to begin with an early configuration to learn from the experience. We wanted to learn not just about the product but about the experience of planning, deployment, and the whole life cycle.
Since the summer of 2022, we now have the final configuration that we are going to deploy in 2500 sites. Of course, you understand that nothing is ever ‘final’, because products are evolving and improving all the time. But we see our configuration as the one that we’re expecting to use to get to mass scale. So far, it’s performing well. We are currently in that moment where we are striving for KPIs that match or exceed the performance we had before. That process will last until roughly March 2023, and then the ramp will start.
JM: That’s impressive! What KPIs are you focused on for this phase of development?
PM: In particular, for the current phase we are measuring radio KPIs, or essentially parameters that ultimately could affect the customer. That’s the focus for our golden cluster, reaching a performance level to deliver equal or better service to our customers without compromise.
Gradually we will be going into more complex areas. The initial deployment in the golden cluster covers cities in the range of 30,000 to 60,000 people. Then gradually we will move to more populated areas, also including massive MIMO.
JM: Super, that’s a sensible sequence. Of course, one critical area of performance is the power efficiency of the RAN. I get a lot of questions about the energy efficiency in baseband processing and the roadmap for Sapphire Rapids. Cristina, how is Intel working to improve energy consumption?
CRISTINA ROGRIGUEZ: Intel is investing in a roadmap that, as Paco says, will make sure there is no compromise on the performance and the service to the users. Our ambitions are to double the performance per Watt in every generation of our CPU products. Essentially, that’s doubling the capacity within the same power envelope.
Intel has multiple avenues to drive the efficiency and performance. For example, in our 4th Gen Intel Xeon processor, we have our new AVX instructions that help with the RAN workload. We also announced recently that our 4th Gen Xeon processor will include Intel vRAN Boost, meaning that we are integrating Layer 1 acceleration.
So, if you look at the solutions that have been deployed so far, we have external accelerators doing specific functionality such as Forward Error Correction. Today, we are simplifying that design and pulling the acceleration on to the CPU die—in the SoC—to include the entire workload. That brings an additional 20% of power savings, on top of other benefits from multiple areas, such as TCO and system simplification.
Then we have advanced observability, also called telemetry, and power management features. As one example: In a recent O-RAN PlugFest test, Vodafone, Intel and other partners were able to deliver 12% end-to-end power savings through the use of core frequency adjustments on the CPUs. We have other sophisticated capabilities as well. There are levers that we call C-states that bring even more capabilities where you can put cores to sleep and bring them back very quickly, saving significant power by automatically adapting to changes in traffic load.
Overall, Intel will double the capacity within the same power envelope, and our ambition is to continue doing that generation over generation. So, in a way, we are creating a new kind of Moore’s Law for vRAN, where you can double the performance per Watt in each generation.
JM: To clarify, the AVX extensions and IPC enhancements are independent of the process node evolution, correct?
CR: There are really three things going on. First, there is the pure architecture with IPC improvements from the micro-architecture improvement. Second, we have integrated acceleration and added AVX extensions to our CPUs. And then in parallel with that, we have the technology node enhancements. They’re separate things, but taken together they lead to great improvements. Why would you do anything else?
JM: Okay. Now, in addition to these hardware changes, what impact will the RIC and xApp or rApp software play in performance improvement?
PM: Last year, Vodafone did a trial of Cohere Technologies’ algorithms that was very positive. We’re now testing this idea in the field because it can have significant benefits in terms of spectral efficiency. Overall, the RIC and app development will take some time, as we will conduct Proof of Concept tests through 2023 and then probably from 2024 it will take off.
JM: Paco, what does this mean in the big picture for a telecom operator? Vodafone has commitments to “net zero carbon emissions” by 2040, and also financial and capacity constraints. Does your strategic decision on Open vRAN match with these other commitments?
PM: Yes, we are a big company and we have different teams. So, we talk to each other, and our Open RAN strategy is an outcome of those discussions. By 2030, we have committed that 30% of our network will follow Open RAN. In fact, as the Head of Open RAN in the company, I am pushing for more than that.
JM: Okay, it’s clear that Vodafone is committed. Do you think that the rest of the telecom players will follow?
PM: One point here is that the separation between Open RAN and Traditional RAN is disappearing. Our forecast is that everything will be open in the future.
JM: That’s also my forecast. Many people focus on the short-term portion of my forecast, which is limited for vRAN and Open RAN due to existing deployments of 5G equipment that won’t be ripped out and replaced. However, I truly believe that when we get to the next major global deployment—whether you call that 5.5G or 6G—it will be “Open Native” and “Cloud Native” from the beginning.
CR: I totally agree and I like the way you are saying it because I believe the network of the future will be virtualized RAN. It has to be virtualized to get the full benefit of Open RAN; in other words, you have to set up the entire telco network as Cloud Native all the way to Layer 1—end to end—to get the biggest benefits. The technology is ready now and the operator benefit is within our grasp.
JM: So, you can try to do Open RAN without virtualization, and you can do vRAN in a proprietary network, but doing both together is important to make it work best?
CR: Yes, to get the benefits of scale and automation, do both together.
JM: Okay, let’s move on to some of the challenges. One area that still needs work is the system integration challenge. NTT DoCoMo has used an interesting approach, where they reduce the integration challenges by creating tag-teams of vendors, matching DU vendors with RU vendors for example to reduce the number of testing combinations.
PM: Yes, that’s a good point. In the future, we can expect to mix and match basically any compliant radio, but right now there’s some integration effort. So, it makes sense to start small. With our selection of partners, that’s what we are doing ourselves. We call this “configurations”, and Configuration 1 is what we have in the UK. We can expect to add more suppliers with every layer as we grow in our deployment, but you need to start in a simple way.
What we’re doing is sharing our experiences and then trying to plan for the future together with others. So, for example, we announced recently a collaboration with DoCoMo precisely to do this. If at some point, we are testing something with Intel, we think it’s for everybody’s benefit that we share the results.
JM: Right. I think that would be very helpful. I suppose that the difficult part will come when some of the configurations move away from x86—for example when some solutions use ARM cores do things get more challenging? Is there any industry collaboration across different silicon architectures to smooth things out?
PM: We have started to look at this and proposed working groups in the Open RAN Alliance. Having the most standardized solution possible is a common goal and nobody wins with fragmentation. So, anything that helps to standardize at the right level falls within the scope of the Alliance, even though it’s difficult to move from competitors to partners.
JM: Yes, that is very hard. I agree that multiple operators will push it, because the operators don’t want to be locked into a network vendor, but they also don’t want to get locked in at another level with a single silicon supplier. Cristina, any comment on collaboration in the O-RAN Alliance?
CR: Yes, let me bring you a little bit of perspective here. Nearly every vRAN deployment today uses the Intel architecture, that’s a fact. And what we’re trying to do is to evolve that architecture and simplify the solution. That’s why we decided to pull in all the acceleration, to put it into a general-purpose processor SoC. By doing that, you just need a server. It doesn’t include an external accelerator of any type, so it’s a server that will be used in vRAN but also used in other workloads. There’s no lock-in. These are common platforms that are understood by all software developers. You get your vRAN, and you get all of your management layers and your entire stack in the Cloud.
The key is to move into the common hardware platform and the hardware/software abstraction so that you can evolve software much faster than the previous paradigm.
JM: I’m wondering about the economics of designing ASICs in the age of Open RAN. As we get into 3 nm silicon and high cost of a tape-out, how does a GPP compare?
CR: Considering that silicon development can cost over $100 million, not to mention mask costs, GPPs allow the industry to leverage economies of scale. Because GPPs are sold in multiple markets, their cost advantages translate into more attractive business models and faster innovation than custom silicon.
JM: What’s next for Vodafone and Intel together? I know you’re collaborating in the Malaga Lab and also some trials in Germany, so can you tell me about what you’re planning there?
PM: Yeah, so our collaboration is now multi-year. I think it was 2015 when we started a joint project with one of the big suppliers in Europe, and it was the output of that work that made me think it was the right time to go with Open RAN. Then we went to Mobile World Congress, we also did a demo using Intel hardware, a very basic LTE cell, but it really helped for everybody to get the idea, and since then we’ve been collaborating in the UK. Our teams work very well together, and I see no other way but for that to be continued. The Malaga lab is one of our main bets in order to continue disaggregating and understanding how we can do better radio. The chips are essential parts for performance and for cost. We started the Malaga lab not just for Vodafone, but as an industry initiative, and we plan to share the outcomes. We already have a few projects we are working on that are initially focused on how we can get the most out of the current Intel roadmap, but also to help with ideas to go from there. We will continue the partnership. It’s essential to us.
CR: I couldn’t have said it better. It’s a fantastic partnership. Vodafone and Intel teams are working well together in multiple countries, and we’re looking forward to continuing this journey. They have been a pioneer on this and understand the issues. Paco wrote this white paper on the challenges of system integration and how the industry should be dealing with this. They are helping the industry to understand the challenges that we have gone through and fixing those challenges. So, I think that white paper, as well as many of their experiences that they have been sharing, is very valuable for the industry.
JM: Let’s move on to our last topic—the future. Paco, can you talk about where you’ll go from here in scaling up to nationwide networks?
PM: Yes, 2030 seems to be so far away, but in reality, to get there with mature services over the network, you need to start deploying much earlier, say in 2025. To do that, you need to have a commercial process closed in 2024 that you are preparing in 2023. Part of that preparation is testing a credible solution. So, we are betting on Open RAN based on solid testing that is happening now. And in the next few months, we will release a few more announcements.