What Data Center-as-a-Service Means for Juniper Networks

The networking giant’s CTO explains its opportunity for a SaaS play and his vision of data center automation.

Yevgeniy Sverdlik, Former Editor-in-Chief

February 2, 2021

When Raj Yavatkar joined VMware four years ago, his new boss, Ray O’Farrell, currently VMware’s CTO who was then running its Software Defined Data Center division, tasked him with helping the company find its place in the new world, dominated by hyperscale cloud platforms.

VMware, which had mushroomed thanks to the success of its server virtualization software in enterprise data centers, was struggling with the next, post-virtualization phase of the cloud revolution. A lot of the enterprise workloads were now going to the cloud providers instead of enterprise data centers.

VMware “did not have a good story around cloud, and they were getting beaten up in the market,” Yavatkar, now CTO at Juniper Networks, told DCK in an interview for The Data Center Podcast.

He is now defining Juniper’s new role in the cloud-dominated world. While there’s no reason for the company to stop selling switches and routers, Yavatkar sees software – specifically Software-as-a-Service – as its main source of growth going forward.

As fellow and senior VP at VMware, he led the effort to automate the experience of using all the disparate technologies the company collectively referred to as Software-Defined Data Center (SDDC). His goal was to enable a user “to bring up the private cloud in any enterprise, any data center, in half a day,” instead of the three months of professional-services engagements it used to take, he said.

Those efforts culminated in what is now called VMware Cloud Foundation, a data center automation platform that’s a core part of VMware’s successful current cloud strategy. It’s the way its enterprise customers deploy private cloud, it’s what runs on top of bare-metal servers inside AWS data centers as VMware Cloud on AWS and on other cloud partners’ platforms.

Yavatkar, formerly an Intel fellow, left VMware in 2018 to spend two years leading a team of engineers who built cloud networking products for Google Cloud Platform customers. When Bikash Koley, former Juniper CTO, left Juniper to take charge of Google Global Networking, Juniper hired Yavatkar to take Koley’s place.

At Juniper, Yavatkar sees opportunity in building software that automates data center network management and makes it application centric. That elevates network management to the point of being able to solve application performance issues – not just network infrastructure issues – and solve them quickly.

Juniper recently acquired Apstra, an intent-based networking company, and Netrounds, whose software continuously probes the network for performance issues, to support that vision.

“When I talk to many cloud majors and data center operators, they all have aspiration to achieve hyperscaler-like operations experience,” Yavatkar said. The hyperscale platforms have small teams manage massive data centers, and when there’s a problem, they can find the root cause and remediate very quickly, “with the SLA of three hours, 12 hours, and 24 hours – not days.”

To give other operators that experience, Juniper is building a data center automation solution that in addition to networking takes into account all other components, including compute, storage, and applications, and applies AI-driven analytics to identify and fix issues.

“We can raise the level of that automation and insights, and that’s my goal,” he said. “I think Juniper can play a role there, and we can change the level of experience people have.”

The big strategic goal is to make Juniper more than a vendor of hardware platforms by adding robust SaaS offerings and the corresponding SaaS revenue. “That’s where the world is going,” Yavatkar said.

Listen to our interview with Raj Yavatkar, CTO of Juniper Networks, in full on The Data Center Podcast. Besides Juniper’s data center strategy, we talked about his time at VMware, the rise of SD-WAN, how the industry should respond to the SolarWinds breach, and what it is about Google that draws all the top networking talent.

Raj Yavatkar, CTO, Juniper Networks, on The Data Center Podcast

Available on Apple Podcasts, Spotify, Google Podcasts, , and wherever else you get your podcasts.

Stream right here:

FULL INTERVIEW TRANSCRIPT

Yevgeniy Sverdlik, EIC, DCK:

I want to kind of summarize your background first. You've spent 10 years at Intel, as an Intel fellow, then you went to VMware, spent four years there in senior executive roles. Then you went to Google and spent two years there working on Google Cloud Networking. And at the start of last year, you became CTO of Juniper. You replaced Bikash Koley. Bikash went back to Google to head Google Global Networking. Google's obviously doing something right attracting top networking talent. So it must be that they are doing something very interesting network-wise that creates the draw. What was it for you?

Raj Yavatkar, CTO, Juniper:

I think when I was at VMware, I decided to go to Google because Google has probably the largest cloud network in the world. Reason for that is, unlike other hyperscalers, they host about eight to nine applications with more than a billion users each on that network because of the infrastructure used. So I started a career with PhD networking. I was associated with some of the internet pioneers, like my advisor [inaudible 00:01:21], Dave Clark and those people. So I always had passion for networking. So at VMware, I was not doing networking. I was working more on hybrid cloud, multi-cloud automation kind of a product. So I was really attracted to go work on a very large scale cloud infrastructure, network infrastructure and product sorts of cloud networking. So that was the draw.

Yevgeniy Sverdlik, EIC, DCK:

So was the challenge of doing something at that scale, which is unique from anything that you can get to do anywhere else.

Raj Yavatkar, CTO, Juniper:

Exactly. Yeah. The other part is also, if you were around '90s, when internet really started exploding, there's a very small set of people around the country who were networking experts. Most of them are in Google today. So all of my colleagues, I know from those days, when I [inaudible 00:02:12], they happen to be there. So that also becomes a nice draw. You go back and see Van Jacobson is there. [inaudible 00:02:19] is there. There are lots of people like that.

Yevgeniy Sverdlik, EIC, DCK:

So all the cool guys are there and you want to be there as well.

Raj Yavatkar, CTO, Juniper:

Exactly.

Yevgeniy Sverdlik, EIC, DCK:

And there are probably a million of these things, but maybe you could tell us some of the biggest things you learned, biggest lessons learned while building cloud networks at that scale.

Raj Yavatkar, CTO, Juniper:

Yeah. So the most important thing is, it's not building the network itself, but the cloud networking requires you to provide the virtual networking infrastructure and build products on top of that, which is a major software development thing. And what I liked about Google is that it's unique in terms of software development methodology. No other company has the methodology that Google has now because a lot of people left Google to do startups that trickling down in the valley. But Google definitely sets a very high bar in terms of how the software development is done and it was a pleasure to learn and work with some very smart engineers.

Yevgeniy Sverdlik, EIC, DCK:

You're talking about kind of the quick release cycle, constantly updating, that kind of thing, CICD.

Raj Yavatkar, CTO, Juniper:

CICD, but also every developer is responsible for testing the code complete. There's no separate quality assurance team. No separate QA team. So every developer is responsible for the quality of the code they develop and the way they get evaluated every performance cycle, is based on code reviews, quality of the code reviews how you did testing, all that stuff. This is all automated in terms of tools.

Yevgeniy Sverdlik, EIC, DCK:

It sounds great. I need to apply that model to editing and writing. If every journalist is responsible for quality of their own article, my life would be so much easier. I'm also going to copy Google. So let's rewind a little bit to your time at VMware. One of the biggest achievements there for you, it appears, was VMware Cloud Foundation. It's a pretty successful piece of technology, pretty fundamental to VMware's business right now. What was your role in creating that platform?

Raj Yavatkar, CTO, Juniper:

That's a good question. So, being an Intel fellow, I was more of an individual contributor leading technology development in networking system architecture, data center networking, data center systems and so on. So when I've been to VMware, I was surprised that VMware was struggling that time. They did not have a good story around cloud and they were getting beaten up in the market by multiple people. They're missing the trend and so on. So my boss there, Ray O'Farrell, who later on became chief development officer and chief technology officer, gave me a simple... He said, "We're a blank sheet. You need to help us figure out what should we do with cloud?" And with that assignment, I started this project from scratch. It went through lots of code names, virtual rack, evo rack, [inaudible 00:05:16], finally VMware Cloud Foundation. And what I did is I went and talked to a bunch of customers to figure out why enterprises are looking at cloud as the model and what do they want to achieve.

Raj Yavatkar, CTO, Juniper:

And one of the observations was that VMware had introduced the concept of software defined data center to bring all of VMware software and compute storage virtualization networking, the login site, via the less operations all. But they're all individual products and it's a very hard to assemble them together to create the software defined data center. So one of the first tasks I took apart is that will automate everything. With a very audacious goal, that one should be able to bring up the private cloud in any enterprise, any data center in half a day, when it was taking three months using professional services at that time. So we assemble a team which focused on day zero, bring up deployment installation, how to make it extremely automated. Same thing for day one and day two operations, badging upgrading, all sorts of lifecycle management, monitoring.

Raj Yavatkar, CTO, Juniper:

And it took us almost two and a half to three years. Two and a half years to get out. And VMware had some really good early adopter customers who acted as early design partners with us. I remember going to Minneapolis Twin Cities and there was 10 large enterprises there. They're all willing to work with you. And that helped me develop this product called VMware Cloud Foundation, which not only became the private cloud product automation, but also became the first product to run in IBM public cloud.

Raj Yavatkar, CTO, Juniper:

We partnered with IBM public cloud than we ever took it to run in AWS. Today it runs in the VMware on AWS. It's a billion dollar revenue. Since it started chipping, in three years reached billion dollar mine. So I'm very proud of that. And it's not just me. There were a bunch of people I've worked with. I am extremely a good collaboration with a lot of people. But I got this unique opportunity thanks to Vee and Pat to really do something from scratch. Almost like a start and bring all the pieces together and focus on high degree of integration automation. I really love that experience.

Yevgeniy Sverdlik, EIC, DCK:

And so there's a lot of pieces, right? And obviously it was very complicated from an organizational standpoint. Can you share maybe also some lessons learned there maybe for your peers or people in similar positions today that are tasked with building maybe a platform in a big company where there's lots of disparate units and every unit has their own product. What are some of the lessons there that you'd like to share?

Raj Yavatkar, CTO, Juniper:

The biggest lesson is that each one of them has their own product roadmap timeline. So if you go and tell them, "Please do something for me to make this year," it falls on the deaf ears for a good reason. Because they have their own priorities they're trying to do. They are rewarded and evaluated for meeting their own targets. So what I did is that I used my resources as a currency to embed my resources in those teams to say that I'll provide you the resource to do some changes. And those people will sit with your engineers. So nobody's putting a tax on you. I had to do that with vSphere, which is most important business VMware has, vSAN, NSX and those people. And that really created a sense of collaboration, co-development and I did not feel so much resistance to change and making changes which makes integration automation easier.

Yevgeniy Sverdlik, EIC, DCK:

Okay. So people don't feel like someone came and said, "Here, do this thing that's going to distract you from your goals."

Raj Yavatkar, CTO, Juniper:

Exactly. Yeah. That's a very important I've seen working with larger companies, right? When you go to other people, it's important to offer them some help rather than impose a burden. Because even if you come top down, your management supports you, that creates a ground level resistance and that's not good. The second thing I learned very quickly is that we were doing iterative development of our bring up and deployment installation, because that was such a complex piece and we're trying to compress the time cycle in terms of how quickly. So getting this in the hands of some early customers and let them try out and give us the feedback, was another good thing that really helped us.

Yevgeniy Sverdlik, EIC, DCK:

Mm-hmm (affirmative). And you're now CTO, in this role for about one year, you were previously pretty senior as well in your last positions. But how has being CTO different from being one or two steps maybe below the corporate hierarchy?

Raj Yavatkar, CTO, Juniper:

So I think I know the most important differences I see is that you have to take on a corporate level view. You don't have to own anything specifically necessarily, but everything you try to look at it's from all broad corporate perspective across the product portfolio, across the thing. Because my boss, Sir Rami, looks at me as a person who is there to help him guide the rest of the people when it comes to technology, engineering, development and so on. So that's a very different role. It's a role of to some extent, internal evangelist, external evangelist. I get called into the accounts to talk to the execs in large accounts that's having some problem. They expect me to act as the ambassador going inside Juniper and solve the problems. So that's very different. That's something very different.

Yevgeniy Sverdlik, EIC, DCK:

So looking at the whole system, you have a control board with lots of buttons, if you press one particular button, how will that affect all the other things?

Raj Yavatkar, CTO, Juniper:

Yeah.

Yevgeniy Sverdlik, EIC, DCK:

So what are Juniper's priorities in the data center market this year? And then also, what are its more strategic goals in the data center longer term, say five years out? Where's the pack going?

Raj Yavatkar, CTO, Juniper:

So the immediate goal is we just completed our requisition of Apstra networks. We did that because to go into data center against our competition. We did not have a fabric management solution when I came on board last year. One of the things I really push for is first, whether we should the organic development of our own fabric management, but it became very clear to meet the market demand and move fast, we need to do inorganic investments. So that's why we evaluated multiple options and chose Apstra networks. They're part of it. They bring very clearly an advantage with respect to close loop automation, with analytics. And then also multi vendor. They support multiple vendors. So this becomes a good incision point, not just in greenfield data centers, but even at brownfield. Where they already have some vendor, we can go in and set ourselves from our fabric perspective while providing the value they provide to close loop automation and analytics. But my ambition for Juniper is to go beyond fabric management and automation.

Raj Yavatkar, CTO, Juniper:

I'm looking at it from data center automation perspective. When I talk to many cloud majors and data center operators, they all have aspiration to achieve hyperscaler like operations experience. They look at these big hyperscalers and say that with small SRE team, they manage the large data center. They don't divide into network, storage, compute. It's all together, right? And they're able to correlate, they are able to do a root cause diagnostics and mediation very quickly with the SLA of three hours to 12 hours and 24 hours. Not days. Everybody wants to be like that. To be able to do that, we need to provide them a data center automation solution which takes into account, not just networking, correlations with all the other workloads, all other loads like compute storage, analytics and provide a level of automation that allows them to achieve that hyperscaler-like operations experience.

Yevgeniy Sverdlik, EIC, DCK:

Hyperscaler-like. And can you maybe flesh that out a little bit? What does it mean to provide management? You were talking about not just from the networking perspective, right? But also compute and storage and everything else. How far reaching do you see Juniper's role there?

Raj Yavatkar, CTO, Juniper:

So our role has to be driven from networking infrastructure perspective, in some cases plugging into third party tools that I used also, but I'll give you a concrete example, right? If you're a data center operator and you see an issue with one offer workloads, it could be that you are using desegregated storage and one of your applications, which is suppose AIML or a high-performance computing application, it's having some performance certainly. Going from there to root cause diagnostics, which might be one of the lead switches link brought to the spine switch. Or maybe the dips buffer switches are having contention for buffers. That level of diagnostics and coordination today takes a lot of time. It's not easy to do. Because how you monitor applications is very different than how you monitor networking traffic. You have to have some correlations.

Raj Yavatkar, CTO, Juniper:

So we acquired a company called Netrounds. They do active network probing so that they can get data about the end to end service level assurance, which can be correlated with workloads. If we are able to do that in real time, then the root cause diagnostics from workload perspective application becomes easier. That requires us to plug into application level monitoring. We don't have to do ourselves application monitoring. But network is essential part of that solution because if you look at any of the problems you see in terms of compute storage interaction or one workload not being able to access another thing, they all go to networking. So we can raise the level of that automation and insights. And that's my goal. I think Juniper can play a role there and we can change the level of experience people have.

Yevgeniy Sverdlik, EIC, DCK:

And from a business perspective, what is the advantage for Juniper in having that level of visibility, that level of management looking at applications?

Raj Yavatkar, CTO, Juniper:

Because now we are not only providing solutions for switch fabrics, like hardware platforms. We start providing SAS based software products, which add value, which are inclusive of our software revenue. And also that's where the world is going. And we'll become one of the major contributors to the efficiency in data centers. And this is also relevant to service providers. If you look at the 5G densification, it's requiring lots of local data centers, small data centers. But they could be in a metro area. It could be as many as a thousand small data centers. Requiring manual operation of those is not possible even for the service provider. They would be looking for automation. This kind of automation can help them. Because what are the local data centers doing? They're doing aggregation of 5g traffic maybe. There might be video traffic. There might be cloud gaming traffic and correlation to any networking issues for quick diagnostics, quick remediation, self-healing is very important for efficiency.

Yevgeniy Sverdlik, EIC, DCK:

And I have to ask your thoughts on the solar winds breach. That's the biggest story right now in the industry. What should be the enterprise IT industry's response? Should everybody just kind of do better at following the best practices, or should there be some big wholesale changes to the way things are done?

Raj Yavatkar, CTO, Juniper:

Yeah, I think it's more than that, right? It's not going to be sufficient to follow because we are increasingly using a lot more open source packages or tools like solar beams or anything because those are necessary to be able to do threat analysis, threat prevention. And when you do that, the source of the software, the supply chain of the software is not secure. It's not guaranteed to be secure, right? People have done that to some extent in manufacturing supply chain. They have very good tracking. They know using blockchain or distributed ledger, you can actually verify at every point your supply chain security.

Raj Yavatkar, CTO, Juniper:

Software supply chain does not have that concept of security. You source a tool. You get patches. You get hot fixes. You upgrade them, but there is no secure mechanism that is deployed using either distributed ledgers or something like that. That level of authentication we only do in the integrity checks. Either you get an image, you can do integrity checks to make sure somebody has not tampered with that. But that is not enough if the source is compromised and you got delivered a image which is already compromised.

Yevgeniy Sverdlik, EIC, DCK:

So you're saying this after our industry should take a few pages out of the hardware industry's playbook in the way they manage the supply chain. Do you see that happening? There is a tendency for people to, when something big happens, big and disruptive, people kind of get energized about changing things and then slowly but surely things kind of snap back into their old patterns. Do you see the solar winds situation as providing the necessary push for that to really happen, for that kind of change?

Raj Yavatkar, CTO, Juniper:

I'm hopeful for two reasons, right? There are two big parties here who could push that change. One is government, because there's a national security at stake. So something will be done from that perspective to start addressing the challenge. This was done long time back in the semiconductor industry, for example, for different reasons. So I think government can play a big role because they themselves have to secure the infrastructure that our government runs on, right? Military, non-military, everything. Second is hyperscalers. It's a competitive advantage if the hyperscaler can assure its customers, "You run solar winds on my platform, I'll make sure that it's secure."

Raj Yavatkar, CTO, Juniper:

How do I do that? Because I work with these people. Because this concept of app store, that has an iPhone, has any phone, has become a powerful tool because that's a mechanism by which you can police and enforce any controls you want. So they could come at it from a competitive perspective and say that if I brought this better than my competition... And the hyperscaler right now, the top three or top five, the competition is fierce, right? Each one of them is growing at 25, 40, 70%, depends on where you are. And there's a lot of business too, but they also a lot of competition because people seem to be choosing to public clouds and sticking to that typically in a multi-cloud environment. So I think those two parts of the ecosystem could make real change happen in my opinion.

Yevgeniy Sverdlik, EIC, DCK:

So some regulation in combination with one of the very basic principles of capitalism is if your competition...

Raj Yavatkar, CTO, Juniper:

And first is not regulation. It's so much about not regulation, but putting controls on what this source and how this source, right?

Yevgeniy Sverdlik, EIC, DCK:

I see.

Raj Yavatkar, CTO, Juniper:

That could be the way. Yeah. Because I'll you an example. All the public key cryptography got a big boost when the military and government sources, NIST and all, started requiring 256-bit keys and stuff like that. And that became [crosstalk 00:21:24].

Yevgeniy Sverdlik, EIC, DCK:

Oh, I see. So not regulation, but actually requiring asking of companies that serve government to follow certain standards. I see.

Raj Yavatkar, CTO, Juniper:

I'm not a fan of regulation I think necessarily. Regulation is good but it limits really the innovation. So you have to be careful.

Yevgeniy Sverdlik, EIC, DCK:

I see. Hybrid cloud. Companies are combining public cloud with on-prem or cloud deployments. You were deeply aware of all those things when you were at VMware, I'm sure. We may see more public cloud infrastructure like Outposts or Azure Stack on premises in the coming years. How has this trend affected Juniper's technology roadmap?

Raj Yavatkar, CTO, Juniper:

So I think all of those cases that you've mentioned, hybrid networking, multi-cloud, require connectivity. And we provide connectivity. So that part is good news. Even if you have AWS Outposts or any product you mentioned that the age of the cloud or on-prem get deployed, it requires networking. That network infrastructure we provide. But more importantly, there's an opportunity for us to do multi-cloud networking or hybrid networking by providing VPC gateways. So you can take your on-prem virtual private cloud and extend it to any public cloud. If you're multi-cloud, you need connection between them. You need transit gateway. You want a transit gateway in a way that you have connectivity, that doesn't cost you. Some of the hyperscalers charge you if you go to the transit gateway. But I would like to provide you with connectivity across multiple cloud, which doesn't require you to pay for the additional. It's the same bits going everywhere.

Raj Yavatkar, CTO, Juniper:

The other part is SD-WAN. SD-WAN is because of the world increase in COVID. SD-WAN is getting even more adoption, right? That's where the campus and branch connectivity happens, added to the public cloud or the third public cloud in a partial mesh or full mesh. That sort of connectivity, where we are bringing a product to the market, through our 128 Technology acquisition which is completely tunnel-less. Doesn't require tunnels. That's a big advantage to start using that as a way to provide hybrid connectivity. And you can combine security with that. So you don't have to have a centralized security sitting in some hyperscaler, which is distributed security to SD-WAN. That every time I do a session with routing, and I go from a branch to campus in a direct connectivity, I can put security controls. All those as opportunities for Juniper without having to compete with hyperscalers, but be more complimentary to them, but still exploit, all these opportunities.

Yevgeniy Sverdlik, EIC, DCK:

Another big trend is big hardware vendors, HP, Dell pushing the hardware as a service idea. Is Juniper also going to eventually try to sell data center switches on demand as a subscription? Is that something you guys are looking at seriously?

Raj Yavatkar, CTO, Juniper:

I don't want to talk about the exact plants, but I tell you make a statement there, which is important. I think more and more people are going to consume infrastructure as a service. And if you look at from between now and 10 years from now, I see almost everything been confused as a service. So one example of that is some of the colo service providers are already starting to do that. They're trying to provide network as a service, compute as a service. That's a leading indication that that's going to happen.

Raj Yavatkar, CTO, Juniper:

When that happens, I think naturally people like us and everybody else, we have to think about how do we operate in that world where we offer infrastructure as a service, but also more importantly, what software we build on top of that and make it available to make it easier to consume any of the infrastructure as a service? I see that as a big opportunity. It's also transformation that industry is going to go through. Whether it happens in five years or 10 years is an open question. Sometimes it might get accelerated. But we have to be prepared for that. And we're definitely getting ready for that.

Yevgeniy Sverdlik, EIC, DCK:

And I want to go back to the Apstra acquisition. We're talking to on January 28th. The Apstra deal was closed just yesterday. At least that's when the press release went out. Apstra is described as an intent-based networking and network automation software company. What are the most important pieces of Apstra's technology for Juniper, and how has Juniper been integrating those pieces into its portfolio?

Raj Yavatkar, CTO, Juniper:

Yeah. So the most important thing they do is closed-loop automation. [inaudible 00:26:06] based thing. So they have so-called blueprints, which allow you to design your fabric based on one of the well-known blueprints. And blueprints come with the notion you have a certain configuration intake. What do you want to achieve? And they continuously verify your running state against that indented state. Any deviation, they can immediately diagnose. Which is a sort of closed-loop automation, which changes the way people think about network fabrics and how they operate. So our first goal is that make sure all of our switches and alters are supported through that fabric management.

Raj Yavatkar, CTO, Juniper:

Second is to take the analytics, take the closed-loop automation and build upon that with our Netrounds active probing capability. We have our own product called HealthBot for analytics. Combining that so that we can provide a lot more value even in the multi-vendor environment with respect to data center automation. So for example, when I'm the priorities would be provide the easiest way and fastest way to do root cause diagnostics. Do that in a way that's useful. Just telling me that one switch cable went bad or when switch fail is not important, if I cannot also have the root cause identification when something else is the first indication of something failing such as my Zoom call failed or the corporate level presentation that was happening has some glitches. Those kinds of things, we have to be able to identify very quickly.

Yevgeniy Sverdlik, EIC, DCK:

I think this was maybe 10 years ago when Juniper kind of took a big stab at creating and selling a fabric. I think it was called QFabric?

Raj Yavatkar, CTO, Juniper:

QFabric, yeah.

Yevgeniy Sverdlik, EIC, DCK:

I noticed there was a lot of press activities around it. They were really pushing it very hard. And I'm wondering today, are any components of those activities are still at Juniper, are still part of Juniper's technology? Are there still echoes of that effort? And then obviously the second part of the question is the very definition of fabric must have changed now. So what Juniper meant by fabric when they were selling QFabric and what you're talking about now are probably very different things.

Raj Yavatkar, CTO, Juniper:

Yeah, I agree. So I think the QFabric insists in terms of QFX switches and some other switching technology, some other buffer switching that we do, which is our biggest differentiator. That is it. What's different is that QFabric was a stateful concept, right? It was a concept of building stateful fabric. Now the fabrics are purely internet based, stateless IP fabrics. That's what has changed. And we have changed with that. We have now the portfolio, which is very modern compatible with what the rest of the world. Advantage of the list thing is that it's much easier to manage, recover from errors and so on.

Raj Yavatkar, CTO, Juniper:

But it also allows to create new problems. Statelessness means that anytime the fabric has issues, you need some good ways to do root cause identification. Because there's no state being maintained anywhere. It's only end to end state. You send in the traffic, it comes out of the other end. It doesn't come out, then you better trace it and find out. That's where the opportunity is with all this closed-loop automation. And that's why Apstra makes a lot of sense to us because they provide that level of automation.

Yevgeniy Sverdlik, EIC, DCK:

And now let's talk a little bit about the pandemic. It's obviously redistributed networking demand geographically speaking. What impact has this workforce redistribution if any, had on the data center networking tech? Has it altered Juniper's data center strategy in any way?

Raj Yavatkar, CTO, Juniper:

Yeah, both. I think first of all, what is altered is that your notion of a branch office or campus has gone to everybody's house. People are sitting at home and working and they want to access both cloud-based SAS services that our corporation might be offering, also be able to come to the cloud data center. So the connectivity and amount of load that you put on your edges in the data center has gone. Right? Second thing that has happened is as, I was saying, using VPN-based tunneling is the old fashioned way. It requires you to really create this VPN sort of mesh if you want to call it.

Raj Yavatkar, CTO, Juniper:

But SD-WAN changes that. So SD-WAN is not just for connecting with your branch offices to campus. It can actually be extended to provide connectivity to remote workers who are doing development. They may have to access databases. They may have to a access big code repository. So that's something has changed in terms of how you think about connectivity to the data center security. Because now security is also distributed. So I think SD-WAN and security together coming in is a big opportunity for Juniper like company.

Yevgeniy Sverdlik, EIC, DCK:

And finally, which three recent technology trends do you think will be the most consequential for data center networking going forward? Things that are happening today that are going to be new trends. Things that are happening in technology that have the most impact on data center networking technology in the longterm.

Raj Yavatkar, CTO, Juniper:

So I think the number one I can definitely, multi-cloud cloud networking. Data center networking is not just restricted to data center. It also needs to provide connectivity to multiple clouds. Because of workloads are distributed, you have microservices based architecture that also may require you to have service meshes that span from a data center to the cloud and so on. So that's definitely a new trend that will have impact with respect to how we think about networking. The second one in a data center space is that the networking is becoming application centric. It's no longer just IP level network. Because a microservices based architecture service meshes is making network more application centric. It's going to the end systems, and that's where you decided on connectivity, your policies, workload placements, distribution, and so on. So that's going to change how we think about networking data center.

Yevgeniy Sverdlik, EIC, DCK:

So that sounds like that also changes the nature of the job of the network admin network manager in the data center. Do you expect there to be a lot of learning curve? What does it mean for people in those roles?

Raj Yavatkar, CTO, Juniper:

So I think that two things happen to them. First of all, lines of business take charge of the networking because they are now consuming it from microservices service mesh point of view. So the service you offer to your lines of businesses changing. It's not driven by you telling them, "This is your connectivity. Use it." They tell you what they want. They are deploying the workload, which requires connectivity between the public cloud and data center and so on. So I think that changes the equation with respect to who has more control. Secondly, what expert is wise also now the traditional networking admin has to now learn more about application level networking. I'm going to be doing load balancing security all at the end system.

Raj Yavatkar, CTO, Juniper:

This no show Smart NICs that everybody's now going for a has two reasons in my opinion. One is, of course x86 scores are very expensive. So Smart NICs allows you to reduce that cost. But more importantly now, switching on like a sidecar proxy, you can decouple all your controls networking such as micro-segmentation, firewall load balancing from the workload independent of them. They don't even have to know, and you can start deploying policies to that. So I think that will change how network administrator now looks at end system. Today typically data center thinks of networking as a fabric, and end systems are managed by somebody else. But with the Smart NICs you're taking the perimeter of the network fabric, extending that to every network end system independent of which hypervisor is running on it or which operating system is running on it. Now you can have uniform policies which span across end systems and network fabric with respect to network security, load balancing, workload placement, affinity policies anti-affinity policies and so on. That's a big change.

Yevgeniy Sverdlik, EIC, DCK:

Okay. So multi-cloud, application level networking, what's what's the third big thing?

Raj Yavatkar, CTO, Juniper:

Third one, I think it's a tricky one because automation has been talked about for a long time. Everybody likes to talk about automation. People have been talking about automation for at least 15 years. I remember there was a IBM autonomics project, which basically said, every system should be autonomic like our human body. We're still not there, but I do strongly believe that we are at the cusp of that trend actually happening, becoming true for multiple reasons. First of all, ML, AI is no longer hype. It's part of everything. You have so much data available and you have computing power to start processing it, applying it, learning from it. So things like predictive analytic, self-diagnosis, self remediation, practical things.

Raj Yavatkar, CTO, Juniper:

And that trend is going to change the way data centers are designed, operated in my opinion. But when I chose the third one, I was in two minds. I also want to mention, which you mentioned earlier, which really want to support, is that data centers help you deal with this new thing happening where everybody wants to consume infrastructure as a service. Which means the data centers might migrate from on-prem to colocations. And if I'm a big IT, I would not have my own data center. Instead, I'll go to colocation providers, say that, "You'll host my data center and provide me compute storage, networking as a service." That's also a big change.

Yevgeniy Sverdlik, EIC, DCK:

It's interesting because, so say maybe five to 10 years ago. Let's just say five. As a service meant AWS, and now the narrative is more as a service doesn't necessarily mean public cloud. It means data center as a service literal data center as a service building and power.

Raj Yavatkar, CTO, Juniper:

And that has an implication, right? You don't have to carry CapEx now on your book. It's only OpEX now. And that OpEx can be directly line of business. So IT's role becomes very different.

About the Author

Subscribe to the Data Center Knowledge Newsletter
Get analysis and expert insight on the latest in data center business and technology delivered to your inbox daily.

You May Also Like