Operators I see as the service providers, the OTTs and the enterprises, OTTs being the datacentre, especially the web-scale datacentre operators. And you’ll see these are fairly commonly known examples of the hot SDN use cases.
But personally where I see the telcos investing are in the public and private clouds, where they want to host enterprise services, starting with elastic cloud provisioning to enable offload into their clouds. And they will either host a private cloud or host a public cloud. And they’re experimenting with this.
And what I’ve seen in the rollout of SDN technology is they will start a new greenfield business with new technology to offer an enterprise-class cloud that’s technically a little bit separate from their current network. And so that’s how they’re getting experience. They’re starting small.
The question of NFV is an interesting one because it starts out as a CapEx saving but it really turns into an OpEx saving. And it will eventually turn into a way of gaining new revenue. And one of the ways of doing that is to open it up to third-party offering of virtualised network functions or VNF. Now some carriers are in the leading edge of this, like NTT. Others like Verizon say I can’t even think about that possibility of opening my network up to third parties offering those services on my network.
The question of DevOps and IT comes up when you start to ask who’s running the network. And it is gradually shifting from the network engineering folks to the IT folks. So as they learn from the IT side of the house how to run software, how to get services and how to have essentially a reliable end-to-end customer experience and quality of experience as opposed to a quality of box performance, they’re going to shift more to doing things in an IT fashion. And there are those that say that the central office is becoming the datacentre.
Finally, telcos, service providers are investing in training of their people because they have a large engineering staff that is doing a lot of things they’re not going to need as much of. AT&T is sending 110,000 of their engineers into AT&T university, offering them short courses in everything about the new technology, focusing, as John Donovan has said, on real-time distributed software.
So the datacentre web-scale operators, where are they investing? They’re investing in white box and bare metal solutions for servers we see in storage and now for networking in terms of switching and in terms of routing. And we’re actually working with some of them to build some product-type examples of that.
They’re also very interested in monitoring and increasing their link utilisation and doing their own home-grown traffic engineering. The most successful deployments we’ve seen of SDN are where they are using it to do in-house traffic engineering that they can do independent of the infrastructure. They can test out new traffic engineering algorithms. They can flip a switch and turn them on and they can achieve very, very high utilisations on their network as a result of their in-house traffic engineering with an SDN underlay.
Enterprises are the slowest ones to get involved in this. And where they’re still doing it in-house, not yet going to cloud, and that’s a different sort of trend, they want to automate things. They’re less concerned with the forwarding play. They want to automate especially things like firewalls. And if they can do automation and orchestration, reduce their OpEx dramatically, that’s where they’re going to see their savings. They’re not so much about creating new services because they’re an in-house operation, but they are looking to reduce their costs.
And enterprise IT has always been a cost centre. Whether it will turn into a profit centre, I don’t know. But it will enable what they do to be more directly tied into the business priorities that the upper management truly is concerned about. So you’re going to see a change in the role of the CIO as a result of SDN into more of a business strategy role.
So where are the vendors investing? So the first bullet is interesting. Semi-proprietary switching and routine. Everyone here is, oh, SDN is the big thing so I have to be on the bandwagon. I have to go to my customers and say, yes, I’ve got SDN solutions.
But if you dig a little bit below the surface, you’ll see there’s still a lot of proprietary technology there. Some of it is necessary because there’s a lot of brownfield installations and their customers cannot change as rapidly as the technology might change. But they’re starting to introduce some of the new technologies. And there are new players. And because of the horizontal layering of networking for the first time, it’s enabled new entrants and players to come to the table with new technologies and new products that break the mould of having everything vertically integrated. It’s like the PC industry in 1981.
There’s been a lot of investment in packet optical integration. The carriers love it. The optical vendors are doing a great job with this. They’re still using some of the legacy protocols through existing equipment, but they’re also putting in new OpenFlow controls down in their control plane. And in a few cases they’re taking it down to the optical element itself.
A lot of vendors working purely in software, and we have networking start-ups now that don’t require $200m in venture capital to build a chipset. And they’re often in some aspect of virtualisation. There’s been a lot of success in virtualising a network, offering an overlay solutions as an easy way to get started. That’s still very popular. It is not the endgame, but it’s created many new opportunities for software companies, both existing and new.
And vendors are investing their people and their money in open source projects, and I’ll say more in a moment. There’s also been an investment in hardware for software-defined networking, which is not what I expected to happen with SDN. And it’s this year the theme is hardware OpenFlow.
We’ve had 1.0 OpenFlow but it had single table and it’s a little bit limited and it’s given good experience. Been a long delay in getting hardware support of multiple tables and very flexible deep packet inspection that’s now coming to the market. We’ve been involved in a lot of aspects of that. We’ll be demonstrating how you’ve got portability between OpenFlow implementations now. And it really does horizontalise the network architecture, abstracts the forwarding plane into this match action paradigm. And it gives a lot more flexibility at how you treat different flows for different revenue and business purposes of your operation.
So we have a variety of silicon solutions now, starting with the traditional merchant silicon ASICs, after the customer ASICs that are dominant in the incumbents, to something new called flexible match-action ASICs. They can look really anywhere in a packet to do a match and then take the prescribed action. No penalty in price performance apparently to the fixed-function ASICs.
A lot of work going on in network processing units (NPUs) and even FPGAs to do, again, deep packet inspection very easily. There’s also a lot going on in CPUs only, mostly at the edge. But if you’re going to have spine switches on top of [X], which is you need hardware processing these packets. But with SDN and the separation of forwarding and control, you can build really high-performance packet processors using a variety of technologies that bring different types of solutions to the customers with which to build best-of-breed solutions.
There’s been some openness in the chipset hardware extraction layers, so the publication of their software development kits. It’s not an open source world there yet. We’re making progress. The open compute project SAI, the switch abstraction interface, is a good step in that direction. And it’s enabling us to essentially free the controllers and the applications from knowing the details of the chip hardware pipelines. Believe it or not, I didn’t think that was required, but it is. And that’s coming along. And a new world of packet programming to really tell a chipset how to configure its pipeline. Very novel stuff. And investments by venture capitalists backing chips.
So open source. A lot of investment by companies. I call this the other OTT, where in this case OTT means opener than thou. Everyone is claiming we’re open. We’re more open than you are. So why is open source so popular? Well, because it’s developmentally efficient for the vendors that don’t want to develop a whole solution themselves because it’s expensive and it’s complicated for certain elements that do not require vendor differentiation. These are some examples. But the network controller space for SDN is a prominent one. It’s complicated. It’s not easy to do by yourself.
And it’s better if you have a community of contributors doing it, each gives a little bit and everyone can take all of it back. Typically it gives you better-quality software. It certainly gives you faster iteration on improvements of the software and feedback from the customers who are also deploying it and contributing to it from their laboratories. And typically if there is a security flaw, it gets fixed very quickly, much faster than proprietary software ever has. And the most successful projects are ones that have a lot of community support.
So there’s a whole lot of stuff in the networking world for open source software. It might make you wonder what’s left. And is there any room for innovation? Any room to make money as a vendor? There’s lots of room. And in fact, what’s left is opportunities to make a real difference for your customers instead of everybody building the same cookie-cutter infrastructure component.
So I’ve given some examples here. The exciting part, of course, is the new services because that means new revenue. Some of the things I mentioned are on here. There’s plenty of opportunity for how do I now abstract the network and get a network map of it? What do I do about building an inventory of all of my components? How can I automate that? How can I turn that into a provision of network functions virtualisation, where I’ve taken stuff out of hardware appliances? I’ve virtualised it.
I’ve put it together and then I’ve used that as a component of a deconstructed OSS. And finally, where is ONF investing as an organisation? We’ve been doing architecture and standards for a long time. We have gotten a lot of traction in what we’re doing in northbound interfaces, because this is where the application writers want to have some commonality. We’re coding them up. And in service chaining, well we work with ETSI and [Avit] to use SDN to facilitate service chaining, which I think is a great step, but only an intermediate temporary step before the full de-whatever of the OSS.
We’ve been working with a number of major operators on how do you get there from a brownfield today to a greenfield tomorrow and to provide carrier-grade services through an SDN underlay. We’re very active investing in OpenFlow. And a lot has gone on in the packet optical integration, also in wireless and in datacentre networking and in the interoperability that has been so elusive and the portability of OpenFlow implementations and hardware.
And we’re investing in open source in key places that help bring it together. Most of our little protocol projects have a coding outcome. We’re doing some selective integration of these major components. We’ll have more to say about that in about a month’s time, in June. And we’re helping to build community, working with these major investments that the other companies are making. I’ve listed them here. We have round tables with the leaders of these. Our members are working in them also.
And we’re trying to bring the best solutions together for rapid deployment and experience by network operators. We’ve been accused of doing PDF-defined networking. Everything we produce is in PDF. And you can’t digest it very well. So we want more consumable artefacts from what we do. So we’re on what we call a journey from PDF to Python. Whether it’s Python or Java, I don’t really care. But how do you get stuff out there? It’s changing the world of what even networking standards are all about.
Dan Pitt Executive Director, Open Networking Foundation