Mark Bergen, Reporter, Bloomberg

Juniper Networks on Automation in the Workplace

Leadership Voices Raj YavatkarNetwork Automation
The still shown features the interviewer and Juniper’s Raj Tavatkar sitting in chairs across from each other with a table in between them with flowers on it. There is a blue sign behind them that has Bloomberg and IBM written across it several times. They appear to be on a stage at a conference.

So long, manual intervention. There’s a new technology in town.

Bloomberg Live aired this interview, with Juniper CTO Raj Tavatkar, from the Intelligent Automation: Creating the Workforce of the Future event. Listen in as Raj discusses how Juniper is leveraging automation to transform the way networks operate and improve outcomes for both the company’s workforce and its customers

Show more

You’ll learn

  • Why Mist is known fondly as “the self-driving network” 

  • How one retailer decreased IT help tickets by a whopping 90 percent 

Who is this for?

Network Professionals Business Leaders


Mark Bergen
Reporter, Bloomberg

Guest speakers

Raj Yavatkar Headshot
Raj Yavatkar
Senior Vice President, Chief Technology Officer, Juniper Networks


0:00 this morning you announced a partnership

0:01 with a company called dragos which is

0:03 cyber security that's right uh could you

0:05 tell us a little bit about that i guess

0:06 i'm also curious sort of

0:08 the old uh adage about buyer bill that

0:10 seems like you've went to a partner

0:12 rather than buy or build uh just a

0:14 little bit about why you made that

0:15 choice yeah so the if you look at the

0:17 industrial control systems or what is

0:19 known as operational technology

0:21 environments like uh electric grid

0:24 those uh environments are subject to

0:27 attack cyber attacks right increasingly

0:29 from east west attacks those called that

0:31 means those originate within the system

0:33 that get compromised

0:35 so we are a leader in providing security

0:37 solutions like firewalls we provide a

0:40 cloud-based security director and all

0:42 that but dragos is a leader in providing

0:44 threat intelligence so if we can get the

0:47 threat intelligence from them

0:49 in real time then we can react in near

0:51 real time to all the cyber attacks and

0:53 protect the environment so that's the

0:55 value proposition that's why we are

0:57 partnering with who what what customers

0:58 are sort of asking for that

1:00 oh many customers as i told you the

1:02 energy great in this country is very

1:04 vulnerable lots of industrial control

1:06 systems in factory automation and all

1:08 they are also vulnerable to that

1:10 traditionally they were air gap systems

1:12 that mean they are separated from the

1:13 rest of the infrastructure but that's

1:14 not true because of internet now so

1:17 those systems are uh increasingly

1:19 vulnerable to attacks from all kinds of

1:21 things those attacks come from within

1:23 that's a real challenge so you're

1:25 traditionally known as you've been kind

1:27 of a customer or sorry a company that's

1:29 worked with the telco industry it sounds

1:31 like when when you joined around 2019

1:33 you've made quite a pivot uh there was

1:36 an acquisition of an ai company called

1:38 mist

1:39 you've moved into services and

1:41 enterprise talk to me a little bit about

1:43 about why uh what that's what's like

1:46 been the biggest challenge for that so

1:49 far so i think juniper traditionally was

1:51 a service provider company people may

1:53 not know approximately about 60 percent

1:55 of the traffic that you see on the

1:57 internet goes to juniper devices and

1:59 systems but the service provider market

2:01 has been flattening or declining

2:04 so the transition we made to

2:05 [Music]

2:07 cloud public cloud providers as well as

2:09 enterprise started with that so we have

2:11 acquired multiple companies starting

2:13 with mist then abstract and 128t the

2:16 idea is to be able to offer both

2:18 software and systems to enterprise

2:21 customers so we can increase that

2:22 revenue could you give some examples of

2:24 where that that pairing has worked it's

2:27 worked well for you multiple places for

2:29 one example is most enterprises manage

2:32 wi-fi networks manually using network

2:34 operation centers on-prem right and that

2:37 is very labor intensive and very slow in

2:40 terms of being able to detect the

2:43 problems and reacting to them

2:46 so what miss brought is that they're the

2:48 first company to do cloud managed uh

2:51 wi-fi uh network management which allows

2:54 them to collect data from all the

2:55 devices from your iphone uh phone ipad

2:58 macbook any of the devices access points

3:01 all that data gets fit to machine

3:03 learning model and machine learning

3:05 models correlate and self-diagnose so

3:07 you can go and start correcting the

3:10 problems before they start showing up as

3:12 alerts in some network operations

3:14 so reduces the amount of manual

3:16 intervention in the network by a

3:18 tremendous amount

3:20 some of our customers for example have

3:22 um one of the largest retailers has

3:25 stated that their troubleshooting

3:26 tickets went down by 90

3:28 so we call it self-driving network it

3:30 tries to self-resolve the problems

3:33 what was that ninety percent over what

3:34 the troubleshooting tickets that used to

3:37 resolve manually now they are getting

3:39 self-resolved by that over like a year

3:41 on year two yeah for three-year period

3:43 yeah yeah yeah and so and you're seeing

3:45 this in in retail as one of their other

3:47 industries retail healthcare education

3:50 lots of universities are using our

3:51 products more and more we are beginning

3:53 to see federal gov

3:55 the federal accounts also beginning to

3:57 see that that's one example the same

3:59 thing with data center we acquired a

4:01 company called abstra that does close

4:03 loop automation of network fabrics and

4:05 data center where the idea is to build a

4:07 model of the network and upper a a a

4:11 prior

4:12 before deploying the network and then

4:14 you constantly monitor it to check

4:16 against the model and if you see any

4:19 diversion from the model you try to

4:20 correct it

4:21 so what's changed starting in march 2020

4:25 and then

4:27 yeah what changed is i think

4:29 they're

4:30 good and bad i think the good thing is

4:32 the demand increased because a lot of

4:33 people started working from home so

4:35 service providers started ordering more

4:37 equipment

4:38 we saw a lot more demand for hybrid

4:41 workforce which means the broadband

4:43 connectivity to home became more mission

4:45 critical required more security so we

4:47 have seen a lot more increase in both

4:49 enterprise

4:50 cells as well as service provider side

4:54 what's been the biggest challenge for

4:55 you you think on the technical side as

4:57 for the pandemic i think the biggest

4:59 challenge has been that when this hybrid

5:02 workforce happened you want to deliver

5:03 the same experience to the home worker

5:06 that you do to the office worker so it's

5:08 no longer the hub and spoke model you

5:11 know that

5:12 the campus is the hub and the spokes

5:14 everything became distributed so that's

5:16 the scale has to increase so i think uh

5:19 anything we do in terms of automation

5:21 helps there

5:22 so you i mean you manage a team but you

5:24 probably can't share how big the team is

5:26 i mean

5:27 so my role is as a chief technology

5:29 officer to drive new products new

5:32 technology strategies so my team is not

5:34 more than 300 people but engineering

5:36 team is much larger so i work closely

5:38 with them yeah i'm curious what how you

5:40 think about the trade-offs and sort of

5:42 if

5:42 if you think about it as a trade-off i'm

5:43 sure sure it is and and sort of the

5:45 traditional networking business and then

5:48 and then some of the advances you're

5:49 doing and on ai and software like how do

5:52 you measure that and allocate yeah i

5:54 think the skill set modernization is the

5:57 one of the challenges right because we

5:58 have to bring more people with

6:00 experience in cloud native architectures

6:02 uh artificial intelligence machine

6:04 learning and that's the transition we

6:05 have been making both organically as

6:08 well as inorganically yeah um you

6:10 mentioned the cloud the the big

6:13 hyperscale cloud companies are customers

6:16 yes

6:17 um they're also

6:19 all of them in amazon microsoft google

6:21 are sort of building their own custom

6:23 silicon or talking about it

6:26 uh and sort of

6:27 investing and sort of owning

6:30 using vertical jargon here but like

6:32 vertical um they're owning a lot more of

6:35 the of everything uh how does that

6:37 change things for for your business so i

6:39 think uh where people are trying to

6:42 build ships is in my opinion commodity

6:44 market which is a very

6:46 very well defined functions where we are

6:48 focused on developing custom chips uh

6:51 which are very energy efficient i mean

6:53 generation to generation we reduce the

6:55 energy consumption of our routers by

6:57 thirty to forty percent they they say

6:59 thirty to forty percent by using custom

7:01 chips which are designed to be very

7:03 highly performant but also very energy

7:06 efficient what's the trade-off there

7:08 so i think uh

7:10 when we build the product if we use uh

7:12 those chips

7:13 we control our destiny right we are not

7:15 only providing performance we can

7:17 provide the highest possibility of the

7:19 performance allowed by the technology

7:21 curve and those products we sell to

7:22 hyperscale yesterday they are not

7:24 building those themselves they're buying

7:25 from us

7:26 what keeps them from building that

7:28 tomorrow i think eventually everyone

7:31 will try to tries to go up the

7:33 technology curve and our our opportunity

7:35 is to stay ahead of it as long as we

7:37 stay ahead of it uh they will not catch

7:39 them i got it i was talking to my

7:41 colleague who covers your industry more

7:43 closely and said well you know the cisco

7:45 is saying kind of the same thing

7:47 i guess i'm curious what you think and

7:48 i'm sure that not just cisco but but all

7:50 the competitors are moving in this

7:51 direction

7:53 um like how do you

7:55 stay out of the curve yeah so i think

7:58 some companies have been traditionally

8:00 focused on enterprise market so for us

8:03 that's a focus shift that we have made

8:06 and we're trying to take advantage of

8:08 our high-end products and bring them to

8:10 the enterprises so that we can

8:12 not only

8:13 grow but grow with differentiation both

8:15 in software and hardware systems that's

8:17 the strategy we are following yeah

8:19 um

8:20 supply chains kind of out of whack

8:22 uh what does that mean for you

8:25 so i think uh uh

8:26 whole industry is facing supply chain

8:28 but so far we have been able to manage

8:30 efficiency of the supply chain by

8:32 diversifying as much as possible so in

8:35 immediately we do see that we'll

8:37 continue to meet our demand

8:40 but our backlog continues to increase

8:42 which is good and bad for future but i

8:45 think uh so far so good

8:47 and you also i saw the most recent

8:49 earnings it seems like you're hiring

8:51 which just seemed to be the opposite

8:52 trend of a lot of companies around here

8:54 yes um do you are you how

8:58 what's your

8:59 outlook on sort of how long that lasts

9:01 is this

9:03 so we are hiding in two places like i

9:05 said in r d we need to hire for new

9:06 skill sets like cloud native

9:08 architecture artificial intelligence

9:10 machine learning that will continue but

9:12 also we are modernizing ourselves first

9:14 by having more in enterprise salesforce

9:16 so because our focus on salesforce as

9:18 long as our growth continues we'll

9:20 continue that path i think yeah i'm

9:22 gonna go back to to just just make sure

9:24 this like the the self self driving

9:26 driving self enough driving network for

9:28 for uh a quick

9:30 second like what um

9:32 are you seeing the sort of customers

9:34 that where where are you seeing like is

9:36 it sort of your existing customers that

9:38 are moving to this is it new customers

9:39 that you've never had before

9:41 uh kind of so when it comes to

9:44 self-driving networks especially in the

9:46 case of mystery one we have taken the

9:47 strategy to apply that principle to

9:50 wi-fi networks access

9:52 like security access and all as well as

9:55 campus branch

9:56 for that we're seeing new customers

9:58 because this is a completely new

9:59 approach where we are displacing

10:01 traditional vendors who are doing manual

10:04 intervention driven network operations

10:07 to automation so that's completely new

10:09 and that's where the growth is coming

10:10 yeah how do you stay up uh on the latest

10:13 in in ai and i know that hiring is

10:15 clearly like a huge challenge like

10:16 everyone's trying to hire ai engineers

10:19 two ways one is that inside juniper we

10:21 created what we call juniper university

10:23 where we brought the experts to start

10:26 building curriculum to

10:27 retrain our engineers and secondly we

10:30 focus on hiring new college graduates

10:32 because if you catch them young and

10:34 early you can develop them into

10:37 more experience not like poaching from

10:39 deep mind or something like that

10:43 yeah

10:45 and but that's the like that's a better

10:46 strategy you think there's like yeah i

10:48 think the better strategies to uh

10:50 invest for long term so i think juniper

10:52 university is a good way to retrain our

10:55 engineers who are hungry to learn new

10:57 skills and get the new college graduates

10:59 yeah i'm a quick one before we jump to q

11:02 a because i'm reluctant i'm not an

11:03 expert in like web 3 and and blockchain

11:06 but i'm sure you're thinking about it

11:07 yes

11:08 where

11:09 what are your thoughts

11:10 what's the how much how much time do you

11:13 spend thinking about it like so i think

11:14 that's one of the my areas is that what

11:17 comes next three to five years we have

11:19 to worry about is the web3

11:21 there's also a lot of hype around it but

11:23 of course the underlying technology so

11:26 called distributed ledger is

11:28 worthwhile it's going to allow us to

11:30 secure supply chains and other parts

11:33 that's what we're looking at how to

11:34 apply it to the places where it's

11:36 designed to be

11:37 uh but not necessarily go by the nft and

11:40 cryptocurrency

11:42 um okay

11:43 that we didn't prepare for my poor

11:44 eyesight

11:46 uh

11:47 how are you first one is how are you

11:48 trying to solve the cybersecurity talent

11:50 gap

11:51 uh with automation i'm not as familiar

11:53 with the cyber security talent gap but i

11:55 imagine you know what that means my i

11:56 think talent gap i think they're

11:57 referring to is that hiring people with

11:59 security expert is also hard because

12:01 there are only

12:03 rsa conference right now is full of them

12:05 you can go steal some of them but uh uh

12:08 i think the

12:09 the question

12:11 really

12:12 leads me to the same answer right that

12:15 by putting more automation that example

12:17 of dragos and juniper partnering today

12:19 will be taking the threat intelligence

12:21 feeding that into machine learning

12:23 models so that we can do threat

12:24 protection that doesn't require as many

12:26 cyber security security experts that you

12:28 would normally get

12:30 uh the next one's really interesting

12:31 into kind of a two part i think um

12:34 so the expansion of the remote workforce

12:37 uh and a lot more

12:39 wi-fi connections and devices

12:41 uh one i guess what does that mean for

12:44 for cyber security and and and two and

12:46 maybe well man that's more important but

12:48 what does it mean for network

12:49 reliability yeah uh and what have you

12:51 seen so far so i think that's what with

12:53 hybrid work model what happened was the

12:56 home became as much a mission critical

12:58 part of the network as your office

13:01 so you have to take the technologies

13:03 like sd-wan and

13:05 also make them available for home juices

13:07 so what we have done with the 128t

13:09 acquisition we combined our security

13:11 portfolio with sd-wan so that for branch

13:14 offices home offices we can provide

13:15 secure access

13:17 and that like do you feel like it's

13:18 straining like how much can you can you

13:21 give a sense about like

13:23 the strain is it sort of like continuing

13:25 this to strain or is it kind of reaching

13:27 i think it's not a strain for us which

13:28 is a it's a clearly growth opportunity

13:31 for us but i think the strain if you use

13:33 the traditional methods of network

13:35 operations if you're going to have a

13:37 bunch of people sitting at a network

13:38 operation center watching monitors

13:40 getting audio visual alerts that's not

13:42 going to scale so by using this ai

13:45 driven self-driving network approach we

13:48 believe that we make the problem simpler

13:50 and

13:50 you can tackle it very easily uh this

13:53 next question is really interesting

13:54 actually um

13:55 what do you think about open source ai

13:57 driven network monitoring solutions uh

14:00 the example was sort of like if if there

14:02 is and maybe it exists i don't know but

14:04 a tensorflow for network security from

14:07 um would that come out from juniper

14:09 yeah so the first of all the open source

14:11 solutions are not a threat

14:13 i really like the open source ecosystem

14:15 because it allows us to contribute but

14:17 also take from the contributions of the

14:19 other and have you been we have been

14:21 very actively uh participate in open

14:23 source but what we'd like to do is

14:25 somebody has to take that solution

14:27 harden it make it

14:29 accessible and usable by customers

14:32 that's what we can do so i think we can

14:34 take advantage of the open source in

14:35 both ways contribute but take it so

14:38 can you give me a sense about how much

14:39 you're contributing now so so for

14:41 example it's the kubernetes is one of

14:43 the most underlying cloud native

14:46 networking architecture from google

14:47 google put in open source we are active

14:49 contributors to that we now have our own

14:52 kubernetes-based cloud networking

14:54 solution that we make it available to

14:56 others uh

14:59 jumping off from web 3 speculation to

15:01 quantum computing speculation uh

15:04 but what worries do you have in a

15:06 potential quantum future about security

15:08 and encryption yeah so the quantum

15:10 computing one of the things we have done

15:12 and with this public information we

15:14 started using it for quantum key

15:16 distribution

15:17 in a secure channel you need a key

15:19 distribution and quantum

15:21 technology computing allows us to make

15:23 it more secure the question refers to

15:25 what happens if you start using quantum

15:27 computers

15:29 to decrypt the data by storing it i

15:31 think there are more post-so-called

15:34 post-quantum encryption algorithms

15:36 already being developed in academia and

15:38 we are looking at some of those to

15:39 collaborate with academia to evolve

15:42 those so by the time the quantum

15:43 computing starts getting applied to uh a

15:46 traditional encryption we should be the

15:48 post quantum era with respect to

15:50 encryption you should be this in the

15:52 post era with respect to encryption

15:54 algorithms and so on the guy because

15:57 right now i think the problem is the

15:58 following right if you

16:00 take the any of the encrypted data store

16:02 it and feed it to any quantum computers

16:04 maybe a day later you will have it

16:05 unencrypted so that's a big thread to

16:08 the traditional encryption algorithms

16:10 yeah so people are developing so-called

16:12 post-quantum encryption algorithms which

16:15 will not be

16:16 um decryptable by quantum computers

16:19 that's where we have to go

16:21 how much i'm i'm a quantum skeptic a

16:23 little bit but because people have been

16:24 talking about it for a long time

16:26 but do you how much of your time do you

16:27 spend thinking about that so we we have

16:30 done a one sort of uh pathfinding

16:32 project only poc where we try to use it

16:34 for key distribution that seems to work

16:37 we are working with the uk's department

16:39 of defense to try it out oh wow so it's

16:41 very early it's early but you have to

16:43 explore that before the things get there

16:45 all right i really appreciate it thanks

16:47 raj for your time thank you everyone for

16:49 very excellent questions

16:51 and i'm going to turn it back over to

16:53 janet

Show more