The Visionaries: A Viewpoint on the Future From Networking Legends | Juniper Global Summit
What’s next for the networking industry?
From the collective movement to the cloud to the importance of security, there’s never been more change in the networking industry than there is today. Moderated by Maribel Lopez, this panel of industry visionaries will help us look a few years into the networking future.
How AI will shape the future of networking
What application-centricity means in a multicloud world
Why unique user experiences are becoming more important
Who is this for?
0:15 -Hi, I’m Maribel Lopez, the founder, and principal analyst at Lopez Research. I’m thrilled to be here today joined
0:22 with such amazing and disparate luminaries. Today we have with us, David Cheriton, Bob Friday, Kireeti Kompella, and Andy Ory.
0:32 Why don’t we get started with introductions? David, tell us a little bit about yourself.
0:37 -Well, let’s see I was a professor at Stanford in computer science and started the first networking course there 40 years ago,
0:46 and since then, I’ve taught networking and distributed systems. I’ve also been involved in founding several
0:54 networking companies and most recently, I was the founder and CEO of Apstra that was doing network
1:01 automation, that got recently acquired by Juniper, so here I am.
1:06 -Wonderful, Bob. -Yes, Bob Friday. Yes, so I started my career in the wireless network back
1:12 in the ‘80s when the FCC came up with the unlicensed spectrum rules. I send them a Christmas card every year, saying thank you.
1:18 This whole career’s dedicated to the FCC and those rules. From there, I went on to basically do a nationwide
1:24 mesh network at Ricochet Metricom long before cellular. After that, I went off and started a company called Airespace,
1:31 which was really focused on enterprise and helping them manage the beginnings of these Wi-Fi networks they were deploying.
1:37 I was at Cisco for eight years, and build the CTO and then went off and started Mist, which was really focused on this cloud-AI paradigm shift.
1:46 From there I sold Mist to Juniper two years ago and now I’m at Juniper and here with you today.
1:52 -Great. Kireeti, share. -Hi, my name is Kireeti Kompella. I started at Juniper 24 years ago,
2:01 in the fifth sub-basement, writing kernel code. I crawled my way up out of there and started working on routing
2:09 protocols going to standards bodies, going to the IETF, got interested in advancing that routing standards,
2:19 but moved on to the last five years, where I’ve been talking about self-driving networks
2:25 and the technologies that we need to get there. Now I’m working on a project that is
2:31 a building block for the self-driving network. Happy to be here. -Thank you. We’re excited to have you.
2:38 Andy, let’s round it out with you. -Yes, I’ve been involved, like Bob, since the ‘80s in communication infrastructure.
2:45 I’m a serial entrepreneur, and I’ve started and successfully grown three different businesses.
2:51 We’ve engaged in deploying products in 110 different countries, and my latest business 128 Technology,
2:58 that I was a co-founder of, is now part of Juniper Networks and it really is a combination of trying to realign the connected world
3:06 from what it was used for, to how it is being used, and anticipating its future use,
3:12 in a way that can impact business outcome to create enterprise value. -Okay, I think that this is great that we have so many people
3:21 representing so many different parts of the stack here. Speaking of the stack, I’ve been in networking
3:26 and telecom and district for a long time. In that space, I think we’ve come to the point where a lot
3:32 of people talk to me and they say, "Hey, isn’t this a commodity? Is there anything really interesting going on with networking?"
3:40 I’d like to pose that question to this panel. Is that true? Do we feel that it’s over in the networking phase?
3:48 What should organizations be doing right now? Maybe we can start with you, Andy.
3:55 -Sure, there’s never been more change in networking than there is today.
4:00 Between the movement to the cloud, the connection of wireless networks, mobility, the importance of security,
4:06 and this whole notion of digital transformation and engagement. We’ve all, unfortunately, lived it over the last 12 months,
4:14 but it is just facilitating massive change. I think that you can look at parts of networking
4:20 that are commoditized and the market values that one way, and you can look at the part of networking that is fundamental
4:26 and strategic to what enterprises are trying to engage in, and that’s highly valued.
4:31 I tend to want to skate to where things are valuable and strategic because I think that means they’re exciting.
4:37 -All right, so you think networking is hot? Anybody not thinking networking is hot? Anybody think it should be different?
4:42 David, what’s your perspective on this? -I think that networking is a little boring at one level,
4:49 And we’re converged around Ethernet. we’re converged around IP, and a lot of stuff has converged compared
4:57 to the diversity of network technologies we had years ago.
5:03 One of the things that has also changed is,
5:08 as Andy mentioned, that the network really is the foundation for digital transformation.
5:14 I like that metaphor in the sense that all the digital transformation sits on the network.
5:19 The metaphor breaks down because this foundation can fall through the ground very quickly.
5:26 To me, one of the exciting pieces is that we’re dealing with,
5:31 not a piece of concrete, but a complex set of interacting components that are
5:37 doing trillion things a second, and when things go wrong, your foundation falls apart.
5:43 To me, the exciting part is making that foundation as reliable as a skyscraper foundation.
5:51 -I love that concept. You need a foundation to build on. It looks like Kireeti [crosstalk]-- -I’m going to pick up on what David just said because
5:57 I’m going to take an analogy with the power networks, the distribution of power to our houses.
6:06 Back about a hundred years ago, there was this big fight about whether we should use DC or AC.
6:15 You can read all about it, but in the end, we did shift to AC. We run 110 volts here, over in Europe there are 220 volts.
6:25 There was a lot of excitement, a lot of stuff going on there, but it’s all settled down.
6:30 The fact that it settled down might mean that it’s a commodity and it’s boring.
6:37 What it does mean for me is, I don’t worry about, "What plugs do I carry?
6:45 Does my thing work on 110 or 220 when I’m traveling?"
6:50 I think it would be great for networking for the end-user to get a little bit more boring,
6:56 where I don’t really worry about how I would get my connectivity when I got there. I just have your big participant
7:02 connectivity. It would be good connectivity. Behind the scenes, there could be a lot of stuff going on.
7:10 What we’re having today is trying to manage solar power, wind power, coal-based power, and manage it in the right way,
7:20 but for the user, I just take my stuff. I don’t think about it. I don’t worry is it 50 Hertz or 110 volts,
7:28 I just plug it in and it works. The one thing you shouldn’t do is try to hammer your plug into a socket of a different kind.
7:35 If you just get the right socket converter, you’re fine. Can we get there with networking?
7:42 We’re far from there now. I think it would be nice for the end-users
7:48 for networking to be a little more boring. -Not boring enough is what I’m hearing.
7:55 -For me, Maribel, when I look at this, I think at this in the context of evolution. If you look over this, my lifetime of the last 30 years
8:02 of working in networking, we’ve gone from internet, no mobility. When I started,
8:08 I was happy to have a 9.8 Kilobit per second internet connection. If you look at evolution,
8:15 the networking is like a living organism, is evolving quicker now than it ever has in the last 100 years.
8:20 We’ve gone from almost no mobility to we’re now almost 70% of all our internet traffic is over some mobile connection.
8:27 If you look where we’re headed in the future, we’re connecting more things. When I look at networking right now, we’re just evolving
8:35 from workloads that were in the enterprise space. All those workloads are now distributed across data centers everywhere.
8:41 If you look in the future right now, networking is going to be evolving faster than what we’ve seen over the last 30 years.
8:47 For me, it’s far from a commodity. It’s an evolving organism that is evolving faster than we’ve ever seen.
8:53 -What I really love about what I’m hearing here is we’ve gotten everything from network being the foundation, which is true,
8:59 to being a living, breathing organism that is constantly evolving, which makes a lot of sense,
9:04 and is what you need for your business right now. It’s never been clearer that agility is the name of the game,
9:10 so creating that foundation that can also be agile is extremely important. I know in several discussions of having each of you in the past,
9:17 we’ve talked a little bit about that agility and that adaptive nature. What should organizations be doing right now, if that’s the case?
9:26 Bob, what do you think people should be doing right now? -For me, in my career vastly,
9:32 the latest transition I’ve seen going on is really this transition from 20 years ago when I was doing networking.
9:39 It was all about helping people manage their devices’ elements. I think right now we’re starting to see this transition
9:46 where it’s really managing this end-to-end experience. It’s no longer good enough to just manage the AP, the router,
9:53 the switch, or the device in your home if you’re a consumer. It’s really, what you care about is, "What is
9:58 my internet connection real experience like?" That’s really the end-to-end user experience.
10:03 When I talk to people, it’s more about connectivity because we’re seeing things get much more distributed.
10:09 The days where everything was self-contained, those days are over with. Now, we have applications in devices running very distributed
10:16 that we’re trying to connect to over the network. Our problem is really, "How do you manage these connections?"
10:21 and not so much about managing the elements in the network. That’s when I talk to people, it’s more about how to deal
10:27 with the complexity of the network that’s growing around you now. -Kireeti, what are you hearing?
10:34 -I think this drives into the previous question.
10:39 If you can make networking boring for the end-user, in the sense that they don’t have to worry about,
10:45 "Which power am I connected to?", to Rob’s point about, we are all mobile,
10:50 what frequency should my mobile connection be connected at?
10:56 If you can take-- Your phone will just work and your connection would be just fine.
11:03 It’ll adapt to the bandwidth that’s available. Behind the scenes, there’s going to be a lot of things happening,
11:09 so that that experience of the user is getting is actually met.
11:15 That’s the dichotomy. For the user, things should be easy, should be maybe boring.
11:25 They should have very little to do. What’s behind that should say, "I’m going to be constantly
11:32 adjusting things so that that user’s experience is great." That constantly adjusting could be done by a human,
11:39 but also could be done with the help of automation. That combination of a human overseeing it,
11:47 automation doing the heavy-lifting, but the end-user getting the experience that they want, if we’re
11:54 in that place, then we’re so much better off than we are today. We’re moving in that direction, but we’re not there yet.
12:02 -Can I jump in here? Slightly, you said what should organizations be doing now?
12:09 I believe we’re facing a bit of a crisis here because there’s not enough
12:16 skilled network operator engineers to operate the growing set of networks.
12:22 There’s 30 to 40 years of evidence that most of the failures are caused by human error.
12:29 As we said, the network is the foundation and we’re still operating networks
12:35 at the level that we used to operate cars when we adjusted the spark advance, if people know what that means.
12:44 We’ve got to get out of this and the answer has to be automation. There’s no other option on the table.
12:51 I think there’s significant progress being made in how we automate networks. This is not your grandmother’s
12:58 network management where all it does is throw up information and you’re supposed to make sense of it.
13:05 This is trying to take over the configuration, the monitoring, the root cause identification of what’s going on with the network.
13:14 I believe we’re at this point where we can automate this and we can dramatically improve the operation of the network,
13:22 and improve the experience of somebody running a network, which right now, is the pretty horrible one because you’re basically
13:30 on call 24/7 to solve inscrutable problems. It’s hard to stay married.
13:38 -If I can add on to that, David, I agree with you. I just think it’s important that we understand what
13:44 it is we’re trying to automate and what our goal is. From some perspective, a network is a cost center,
13:49 and when asked how is the network performing for the organization, you can look at link status, you can look it up-time, and you can say here’s how it’s performing.
13:56 I think that all of us on this bridge think that networks can also be strategic, that can be value drivers for an organization.
14:04 For that to be the case, networks need to be able to do something they can’t do today. They need to be able to understand who is
14:10 using it and what they’re trying to accomplish, and what else they could be doing that would help the business
14:16 strategically become more competitive, more profitable, more valuable. I think that as we automate, it’s important
14:24 that we capture or encompass what the strategic role of the network could be and what the business is trying to accomplish.
14:32 -I think we’re starting to hear a lot about this in the telecom sector where they’ve been talking about
14:38 network embedded services, but it looked like David wanted to follow up on that point.
14:43 -Well, I just want to say the I agree with Andy, but I hope we can all appreciate
14:51 that when you add these extra requirements on the network, you’re complicating it further.
14:57 When you say, "Understanding what applications are doing," if you understand it wrong you can make things worse.
15:05 I think that these are double-edged swords and we have to make
15:11 sure automation is swinging the sword in the right direction. Many of the things we might talk about of automation
15:19 may not be completely ready to deploy today, but what I see is a lot of organizations have
15:26 just not prepared themselves to automate. They are not thinking about, "What should
15:32 we be automating, and what tools are available, what do we expect is going to be feasible down the line, and
15:38 how do we get from here to there so we’re not blindsided when this technology is really taking off and our competitors are using it?"
15:49 -Maribel, for those who know me, for me, automation, this AI/ML,
15:54 is really the next ultimate step in this automation phase. If you look at cars being able to self-drive or anything,
16:02 that same concept applies to networking so where we can actually start building things that can actually do things on par with humans.
16:08 Automation is really just that concept. AI/ML is just really taking that automation to the next level.
16:14 Now we’re going to start doing things on par with network domain experts. I think that is the next part where we’re headed is,
16:21 "How do we bring that automation, especially with the complexity of the networking we’re starting to see around us?"
16:26 As Andy points out, these networks are becoming much more strategic to businesses. In our early days,
16:33 a business had all its network inside its firewall. Nowadays, if the internet goes down, their business stops.
16:39 That internet connectivity, wherever these workloads are running, they have to keep that networking up and running
16:44 for their business to stay up and running nowadays. That’s the complexity that AI/ML is going to help start dealing with more and more now.
16:53 -I think that gets us to the next step, whereby not only controlling the complexity within the network,
17:01 we can actually take the complexity out of other systems, starting with humans who are using this network.
17:08 Someone saying, "Oh, I wish I had gotten that higher tier of bandwidth because I’m not getting the connectivity that I need."
17:18 Instead, the network realizes this person is using an app
17:23 that needs more bandwidth and makes more bandwidth available.
17:29 Response to not even the stated intent but the real intent to making apps easier
17:38 to write because apps today have to keep looking at the network and think, "What do I need to do in the network to keep up?
17:45 Oh, I’m seeing that there’s some packet drops so I’m going to adjust. You have this variable encoding of video, and I’m going
17:54 to adjust in real-time to what’s going on in the network." If the network could do that response, then the app can be simpler.
18:02 By taking on that complexity and solving it in the network,
18:07 the complexity for the user trying to understand all the different knobs that they have to be able to deal
18:13 with changes in the network or in the apps, that goes away and we move that to the layer where it really belongs.
18:21 The network has to be ready to say, "I understand what’s going on and I can respond." David’s point is very valid.
18:31 If you start doing a lot of automation, it can cut both ways. If it cuts in this direction and cuts the complexity of the users
18:40 of the network, whether they’re human or applications, it will move the industry forward at huge steps.
18:48 -I love where everybody’s going with this because we’re looking at automation, we’re looking at AI,
18:53 we’re starting to tee up what the future is. I think both Bob and Kireeti, you started to tee up what the future is.
18:59 I’m going to ask you to go one click deeper. Maybe since you were already going, Kireeti.
19:04 Your role is interesting because you look at the 3, 5, 10. When you’re looking at the 5 and the 10,
19:11 what are you thinking about? What do you think the audience should be thinking about? -Bob just mentioned it quickly, this notion of self-driving.
19:20 Today, we’re looking at a lot of self-driving cars. We started back in 2004, with a challenge,
19:28 a grand challenge from DARPA, and said, "Can we build a car that completely runs itself?" At that time, it was, "Wow, you guys are completely nuts."
19:36 Here we are on the cusp of it. It’s still not there, which means that the last bit is still difficult.
19:44 The idea that you have a self-driving car was going to change the automobile industry so completely.
19:52 I think that we can do the same with networks, and we can start in places where it’s a little easier and say,
19:59 "Can I build a network that will do everything that typically humans do to manage that network,
20:06 to bring up new services, to recognize. the customers and make sure that they’re getting what they need?"
20:12 Then maybe go beyond that, anticipate, and do that in an environment that’s relatively simple,
20:20 but then do that across the entire network. This idea that you have an effort that completely runs itself,
20:27 maybe the same way that 10 years ago, 15 years ago,
20:32 the idea of a car that completely ran itself would have been nuts. I think that’s very real,
20:37 and that’s a tangible target that we could go after.
20:42 We are in places, but there are a lot of missing technologies.
20:48 Just as in self-driving cars, there’s this big question of,
20:54 "Do I use cameras, stereo cameras, or do I use LIDAR?" In some driving networks, we have the same question.
21:02 "What is the best source of telemetry because if I don’t have the right telemetry, at the right time with the right quality,
21:08 I can’t make good decisions." We are walking this path, but I think the end of it,
21:16 at least for me, is a network that runs itself. If you go back to the original question of commodity,
21:23 well, from the point of view of the user of the network, if the network is just running by itself and
21:30 giving me what I need, the network itself is not commoditized, but from the user’s point of view, it’s really easy.
21:38 Maybe that’s where you should go. -We talked about this whole--
21:43 People have a lot of data but isn’t the right data. Maybe there’s a role for you to help them
21:49 get to more of the root cause of things versus the myriad of alarms and alerts and data
21:57 and whatever that we were talking about. Bob, I don’t know where you wanted to go with it. -I was going to go with, I make a barrel of wine every year.
22:05 Good wine starts with good grapes. Good AI starts with good data. It’s all about data. I mean that’s the first stepping stone to getting there.
22:13 -[crosstalk] too. Yes. -But I think-- -I can keep you there if you give me the line.
22:20 -I think there’s a challenge here which is that,
22:25 to me, the worst-case scenario that you want to deal with in automation is when something goes
22:32 wrong, you want to identify that as root cause, identify what is the root of the problem as quickly as possible.
22:40 What I see right now is, these people are collecting large amounts of data and then thinking they’re going to process it later and that,
22:47 by definition, pushes it further down the pike, so to speak.
22:52 To me, that’s where you want to turn this whole process around and say,
22:57 "I want to have a diagnostic engine that’s processing the data in real-time, and the diagnostic engine needs to know what it needs
23:05 to know, and that drives the data collection." The reality is that data collection in a network
23:13 is expensive and, of course, the time you’re going to need it most is when things are under peak load,
23:20 and that’s also the time where you’re most likely to have failures, and you’re most likely to have the most expensive failures.
23:28 Being highly selective there is going to be an extremely valuable direction to take things.
23:36 That’s where I’ve been interested in more of the medical diagnosis kind of approach
23:41 to root cause identification where you know what symptoms you’re looking for
23:47 to identify certain problems, so you know what telemetry you need to get. In some cases, you can collect the telemetry on-demand
23:56 once you’ve narrowed it down just like doctors have to do. They don’t get to perform every medical test in the book
24:02 before they do diagnosis. -My only point though is that I think that automation is incredibly
24:10 important, making things simple is incredibly important but, again, it’s only half of the equation.
24:15 You could make things cost less, you could make things appear with a smaller footprint, but if it’s not driving value,
24:23 it’s purely a cost-centric model. I think that many folks who are watching this would be very happy to pay twice as much as what they’re paying
24:31 right now for their network and hire more people if they could if they can create five times the enterprise value.
24:37 I do think that the business outcome aspect is really important to make sure it’s a focus on simplicity and automation.
24:46 -I agree with Andy. Ultimately this is all about trying to make sure that businesses can get their job done.
24:53 I would say, back to the data telemetry question, when I started Mist, it was actually kind of inspired by the Watson playing
25:00 Jeopardy, being able to answer questions on par as humans.
25:06 Like I told you, I make a barrel of wine, so the old saying is, "Great wine starts with great grapes."
25:12 Great AI starts with great data. I look at the problem right now, and I agree with David on the telemetry thing.
25:18 When we started Mist, it was all about making sure we built an architecture that could process data in real-time,
25:24 and that was the big difference from what I did 20 years ago, is we really didn’t have data pipelines.
25:29 We took data and we stored it away into some database. Nowadays, you want to process that data in real-time as it comes in.
25:35 The question is, what do you have to do with that data to get to that root cause? This where I say it’s like the Jeopardy thing.
25:44 Depending what questions or what root cause you’re trying to answer, it takes a whole bunch of different algorithms.
25:50 That self-driving car is not one magical algorithm that drives that car. It’s a whole bunch of complex systems that is driving that car.
25:56 For me, that’s where I think we’re headed right now, is basically leveraging these AI/ML algorithms
26:01 that lets us do things that we couldn’t do 20 years ago. Twenty years ago, I couldn’t do anomaly detection because we didn’t have the fancy math.
26:09 Nowadays, we can do the fancy math that makes anomaly detection actually really work with low false positives.
26:15 I think that’s where we’re headed right now with trying to be able to process data real-time, and it does start with the data.
26:21 You got to have the right data to actually answer the question. -Okay, so I wanted to see if we can stretch it just a bit down the road.
26:29 Andy, any thoughts on what’s next after that?
26:35 -Fundamentally, you’re almost asking what’s the next disruption as opposed to what’s the next innovation?
26:40 I think that we’re all talking about complexity, and as systems become complicated, innovations are harder and matter less.
26:49 Whereas a disruption resets the system, making it inherently less complicated and allows for a faster pace of subsequent innovations, in my perspective.
26:58 I happen to think that the anticipatory side-- When I think of machine learning, I think of adaptation,
27:03 but when I think of AI, I think of anticipation. I think that the extent to which we could really create a robust
27:10 anticipatory paradigm, we would see real subsequent innovation.
27:16 I think that it’s AI in its truest sense, something that doesn’t have a quantifiable ROI in terms of taking
27:24 our trouble tickets or mean time to resolutions to zero, but something that would actually provide
27:30 insight to us that is completely different from the kind of insight we’re getting today
27:35 from our complex systems. -I call this right time experiences, the right information,
27:42 to the right person or service, at the right moment. I think we’ve been moving down that path for a while,
27:48 but to Bob and David and Kireeti’s point, we’ve had some interesting technology steps we had to make along the way to get there.
27:56 Computing has obviously gotten better, we’ve moved into the AI era with GPUs and virtualization.
28:03 Is there anything else we should be thinking about in computing? -Well, the idea that Andy just articulated,
28:10 you could take that further. A lot of people talk about intent-based
28:15 networking, and so you have a stated intent, and then you’re going to move the network to meet that intent.
28:25 When you get to the anticipatory stuff, you have either an unstated intent, so don’t do what I say, do what I’m thinking.
28:36 You have to tease out what is it that this person was trying to do, and then prepare the network for that.
28:44 Or you might have an intent that is a repeating intent, and it might not be repeating in time,
28:50 but there’s some triggers that say, this is what they need. If you can get to the point where the network can adjust to what
28:59 is needed without that thing that’s needed being explicitly stated, then, again, you’ve really moved the network industry far.
29:10 Again, this comes back to, "Is the customer happy?" If the customer says, "Wow, this is almost reading
29:17 my mind and preparing the network for what I’m doing, or just making my life easier,"
29:25 then the relationship between that user and the network becomes so much better.
29:33 -I don’t know if it’s bad if I can disagree with some of my panelists here.
29:39 I’ve been involved with networking for 40 years, I hate to admit, and there’s always been these proposals
29:45 to add various functionality to networks, and yet I don’t think they’ve really gone any place. We’re stuck with it.
29:53 Just basically Ethernet packets flowing around perhaps dressed up with IP Clos on them and so on, and yet,
30:00 you come down to this aspect that, in terms of business value, the business value is that it’s there and you can depend on it.
30:09 We talk about this as a foundation, but it’s this whole collection of complex pieces that are operating with trillions of events a second.
30:18 To me, we’ve got a huge task just to get it to the point that it’s a solid foundation.
30:27 The solution over the last 40 years to applications
30:32 and so on is just to throw more bandwidth at it. When I got in this business, there was kilobit networking,
30:39 we went to 3-megabit Ethernet, to 10-megabit Ethernet, to 100-megabit Ethernet,
30:45 a gigabit, 100-gigabit, on and on and on. I remember some optical guy telling me, "Well,
30:52 people think that optical has got infinite bandwidth, but actually it’s only about 30 terabits of usable bandwidth."
31:01 We got lots of space to grow capacity, what we can’t tolerate is when it just comes to a halt,
31:08 or even what calls a great failure. Now, I think over the next five years, it’s going to be
31:14 transformation, being able to identify root causes of faults.
31:19 What that’s going to do is bring down the mean time to repair significantly, but the remaining tall pole in the tent is going
31:27 to be, "How long does it take to repair something?" I think once we brought this tall pole down, we’re
31:33 going to be looking at techniques to build in redundancy in the networks so that we can adapt
31:39 in real-time, that is, we can auto remediate. I think that’s when we get into a foundation
31:45 that deserves the name foundation. -Well, I think that’s a great point to close on,
31:51 creating a foundation that is rock solid and will deliver business value for you going forward, trying
31:58 to evolve it so that’s contextual adaptive learning, and predictive but once you’ve created that solid foundation.
32:04 Thank you so much for your time and attention, and I’m sure all the panelists are thrilled
32:10 that you shared this time with us here today. Thank you. [music]