Najam Ahmad, Director of Infrastructure with Facebook, hosted a Keynote panel titled “Opening Up Network Hardware” during Day Two of Open Compute Project Summit V. His guests on stage were: JR Rivers (Co-founder and CEO, Cumulus Networks), Martin Casado (Chief Architect, Networking, VMware), Matthew Liste (Managing Director, Core Platform Engineering, Goldman Sachs) and Dave Maltz (Partner Development Manager, Microsoft).

They were introduced on stage by Frank Frankovsky, who stated that “Network has been missing from the OCP for a long time, but in the last 9 months we’ve spent a lot of time investing in networking technologies in OCP.”

“The network industry, especially in the data center world, is going through a massive transition,” Najam Ahmad declared opening the session. “What does the network look like in a three-to-five year frame?”

“Microsoft’s data centers are very large and scale is our biggest challenge that we face,”said Maltz. “We have a million servers now and the number is growing rapidly. The scale comes at a lot of costs. We will be looking at technologies that will help us reduce the fiber-plant, get more information out of each set of fibers. We’ll be looking at ways to speed the scale of our deployments; we have to be able to get devices out, get them deployed and brought up quickly, finding the errors in the cabling and get them fixed as quick as possible. That’s where the software management comes in. You need the ability to manage the switches the way we manage servers today.”

“We have a lot of great infrastructure out there that is enabling us to deploy and update portions of the services on a rapid basis. We need to update the management software for our network fabric at the same kind of speed and agility,” stated Maltz.

“At Goldman Sachs we have slightly different challenges,” Liste shared. “We scale as well, but not to the degree that Microsoft described. We certainly have a lot of complexity in our environment that we built over the past 30 years. Our network has certain degrees of technology in it. We ended up with many different management plains. Then there are issues regarding the security (pretty complex policy management). But building common management infrastructure allows us to get rid of all this complexity,” believes Liste.

“Whenever we talk about the new network you talk about this existential threat that is out there, or the existential change, but if we view the industry like Darwin has already spoken, like it’s already played out, but only in a certain market segment, now we have to find out how we bring the rest of the world along,” said Casado.

• The right migration path

“The biggest trend that happened over the last decade in the data center: the physical network became very simple and functionality migrated to the software at the edge,” said Casado. “Nowadays operations, capital expenditures and innovation speeds are better, this is a great model for building a data center. As the rest of the world catches on and tries to achieve these kind of gains, there’s the question: ‘What’s the right migration path?’

“In many ways, if you have new applications, you put the software in the applications. If you control the platform, the platform itself will assume some of these, and for others there are software layers that are emerging and basically do the arbitrage between the old way the network looks, and the new way that are very simple and provides that functionality,” Casado added.

“We know what the future of networks looks like, we have these proof points, we can see how awesome they are, and now it’s coming with these migration paths. For the enterprise is ‘How you get these networks that are much simpler to look like these old networks but provide the same benefits?’ I do believe it’s software at the edge to do it,” said Casado.

“The problem seems to be the complexity, the manageability and the visibility. What should be do next?” asked Ahmad.

“At the fundamental layer, if someone puts a piece of hardware into the data center, they should own that hardware and they should be able to put whatever software runs on that platform,” said Rivers. “It sounds like a simple statement, but it’s not the way it works today in the industry. Most of the mega-scales all work that way, which gives them the ability to optimize where they position functionality and how interfaces work for operation. That operation efficiency is really what drives the bottom line: why do I make this decision? How do I make the cost of an application go down? Some of it is capacity, some affordability, and some is the ease-of-use,” believes Rivers.

“I always hear ‘put it in software’,” said Ahmad. “Big companies versus small companies – who can afford doing that?”

“If you look how compute evolved and migrated to an open architecture that allowed more flexibility, we believe it’s important to achieve the same in networking,” added Liste.

“I think I was very naive early on. I thought it was really hard to change hardware. Seven years later I realized that probably the most inflexible thing in the industry is the human brain,” joked Casado.

“It is incumbent on us as a community to create solutions that look like the solutions that were used before. If you don’t do that, the tooling and the applications don’t work and people do not know how to deal with it. The delta we have to bridge as a vendor community to help those that are not used to these new models of doing things. Many of the leading guys can build these things on their own and rebuild their applications. Down market, you have to have more tools to help people deal with these things,” Casado added.

Referring to Frank’s keynote, Rivers pointed out that Open Compute is striving to make many pathways to get from a technology provider into the customers hands – what you see is “solutions tailor-fit for different segments of the market,” he clarified.

“What are the specifics? What can OCP do to build this ecosystem? Ahmad inquired.

• Standardization across the platforms

“We need to run services on the network devices themselves, and we need the ability to manage those services so that when they fail – which they will – we can use all our service health monitoring and management to get the networking back to healthiness,” explained Maltz. “Through OCP, I hope we can get some sort of standardization across these platforms so that, as hardware evolves generation to generation, and from device vendor to device vendor, we can use the skills and the tooling that we built in those services that are running on our devices,” said Maltz.

“Does that point to an operating system or a common platform?” asked the moderator.

“The term used is ‘common extraction layer’ and it’s implemented through an operating system, through an API. When I want to execute a service, something as simple as creating a VLAN, I can do so consistently across different hardware layers and, as services evolve, we end up in a place where we can write very interesting applications that control end-to-end because they have a common entry point into the network across many different devices,” answered Liste.

Rivers commented that Liste’s standardization epitomized Casado’s previous statement about the inflexibility of the human brain: “You talked about multiple suppliers giving you the same common API; in the existing networking world, switches and routers are appliances.If you buy from vendor X, you get stuck to their view of the world. If you just bought that hardware, and you can put any software on it, you don’t actually need the normalization that you were talking about,” he declared.

“That’s the value that OCP provides,” agreed Ahmad. “Specific solution providers can provide the best solution possible, at the right prices, for solving a specific problem, and you don’t have to buy a vertically integrated solution set, that may or may not be what you want exactly.”

“Five years ago I was more pessimistic about the networking industry,” admitted Casado. “But now I am so optimistic. We’ve come far enough along, that it’s more about education and pointing out the examples than it is about hardwork. I believe the industry is going to do the right thing. All the forces there are going to come to play, and we are going to end up with better networks. Education is very important to increase the adoption.”

“There’s a large talent pool of people out there who are excellent developers and who understand networking and networking concepts and fundamentals. The intersection of that is still too small today; we have a hard time finding great people with those intersecting talents, but that pool is growing. You can often pair great developers with great network engineers to get a good team,” explained Maltz they way Microsoft’s team gets productive.

• The margins in mystery

“There’s margin in mystery,” ended Frank Frankovsky. “I agree with Martin Casado that, the more we demystify things and educate, the more we can look for those margins. I am not just talking about profit margins; there’s performance margins, operations efficiency margins and we need to look for those margins everywhere around us, challenge us, and figure out how to make it better.”