No-one could accuse Marvell Technology Inc. of not taking cloud infrastructure seriously.
“At Marvell, we believe that infrastructure powers progress,” blares its Web site. And not just any infrastructure; over the last few years the company has focused its efforts on setting out its stall with a rich array of cloud-optimized silicon, as well as plopping down $11 billion to purchase two cloud and cloud-native 5G companies: Innovium and Inphi.
The strategy is working, especially in the virtual radio access network (vRAN) space, where the company is making a name for itself with specialist accelerator hardware that boosts cloud performance.
Raghib Hussain, president of products and technologies, oversees the company’s long-term technology vision and strategy efforts. I sat down with him to discuss the state of the state of cloud infrastructure.
Steve Saunders: Where are we in the development of cloud?
Raghib Hussain: I would say we're still in the early days. Back in 2000, what people were saying about cloud was completely different. The cloud was a bunch of servers put together and everything was done in software. And, frankly, that was very expensive. So in the next decade people started to ask, ok, what can we move into hardware? What can we optimize? So, a lot of functions moved into this hardware abstraction layer, so that the cloud providers could free up the actual CPU, [and] rent it to other people as a resource. As cloud moved from proof-of-concept into a real business, there was a need to optimize it for the power and performance, while providing the same functionality that you could do with software.
Saunders: So, it's a false assumption that everything would be better if it ran in software?
Hussain: It might be easier to run, but it may not provide the solution that you're looking for. In theory, you can say, ‘I have a whole data centre, I can just keep expanding,’ and that’s true but it will cost you more, and latency may be an issue, unless you enable those servers with specialized hardware. For the cloud dream to become reality, the industry has to transition from the early experiments to be able to provide all features at parity.
Saunders: And there are a lot of features?
Hussain: There are. The features in classic RAN equipment have been developed over 30 or 40 years, right? So there are a lot of details, and all those things need to move into this model of virtualized networking.
Saunders: How is that going?
Hussain: Unfortunately, I'm not aware of any company today that can say ‘Hey, I am the RAN software provider, and I have as good of a quality and features as the Nokias and Ericssons of the world.” There are start-ups that have something, but, the operators trying to use it [are saying that it’s] not even close. It's going to take a lot of investment from the overall ecosystem to get to that level. The question is, who is going to invest for that, right? Are operators going to invest? I'm not so sure.
Saunders: Most operators only spend about 1% of their revenues on R&D, so it seems unlikely that they will be the ones to take that on.
Hussain: That’s the problem. They have a capex limitation, and they would rather use that [money] to deploy more cell sites, rather than trying to do their own innovation, right? So, will the cloud players do it? Probably not. They are in the business of providing the platform, 100%. So, that leaves the only people remaining who are actually doing some work on that side as the same traditional OEMs, because it is easier for them to move their traditional software into the cloud than for someone else to invent it.
Saunders: Why is cloud-native 5G important to mobile carriers?
Hussain: The [mobile] operators all want to drive innovation. The traditional deployment model involved going through an OEM, and there are really only a handful of companies that can do that. But with [open-RAN] or vRAN, you have many, many suppliers of the various components of the solution, and they should all be able to work together to drive the vision. So, that's the high-level view. But there's another piece here, which is that, historically, the mobile operators have had to build the full infrastructure themselves wherever they wanted to offer service. But cloud and edge cloud are expanding everywhere. By leveraging that infrastructure to run mobile services, it lets them reduce the cost of deployment and operations. Of course, as you know, there's always a disconnect between the marketing and the reality on the ground....
Saunders: Let's talk about that [disconnect]. Open-RAN (O-RAN) and centralized-RAN (C-RAN) seem well established. What about C-RAN?
Hussain: C-RAN was initially driven by the idea that everything can run in [centralized] software in the cloud. That was the dream, but people worked out early on that it doesn't service the need, particularly with the performance of latency, and especially with 5G applications. You just need to look at how data centers are evolving, they are not always centralized anymore. Microsoft, Google, Amazon all have edge data centres now. Amazon Outpost is a perfect example. It's really a data center in a box. And it lets you bring the application all the way out to the edge of the network, and a lot of things can be serviced out there. Because not every service request has to go to the data centre.
Saunders: Is that's useful in 5G implementations?
Hussain: Oh, it is very much applicable to 5G. If you look at the example of a rural deployment, initially you may not have the user numbers to justify a large data centre installation. So this allows you to start with a very small edge data centre and scale up with the need. Or, in a factory setting you might start with a bunch of these boxes, whereas a small shop, a Starbucks or whatever, just gets one. And if they are 5G capable, an operator can rent that space in that edge data centre to deploy 5G services to consumers, or, in a factory, they can provide the services like our private 5G, where the private network is still connected to the operator network, and they are able to manage it and provide updates and so on.
Saunders: Are you happy with Marvell’s progress?
Hussain: Five years ago, Marvell decided to really focus on data infrastructure. And we have made a lot of organic and inorganic investments so that we can cover the entire life cycle of the data — how it is stored, secured, processed and moved around. If you just look at security, for example, nine out of top 10 firewall appliances all use our processors. So, yes, I'm happy with the platform.
Want to learn more about data strategies for the cloud? Register for our Cloud Data Center Strategies virtual event here.