Private Data Center as Easy as Cloud With Juniper Apstra

0:09 um you may know me as more most recently

0:12 as the uh founder of Abra uh but given

0:16 that recently I got an an expanded role

0:18 focusing on the data center all of the

0:20 data center products for Juniper what I

0:22 thought is talk more about the data

0:25 center and uh you know the the all of

0:28 the things we're bringing as in Juniper

0:30 to the table when it comes to uh really

0:33 solving problems for our data center uh

0:37 customers and so that's going to be the

0:39 focus today and uh we're going to also

0:41 showcase a bunch of those capabilities

0:44 uh you know throughout the session and

0:46 so but before I jump in uh what I

0:49 thought I'd do is kind of step back and

0:51 talk about the evolution of the data

0:54 center and coincidentally I'm starting

0:56 in September of 2 of 1995 which is when

0:59 actually I came to the US uh to attend

1:02 Stanford and um that was coincidentally

1:05 the same uh month where Netscape

1:08 actually went IPO if you recall uh at

1:11 the time if you were there uh it was a

1:14 crazy time where every company was kind

1:16 of rushing to go online um you know

1:19 websites were U Were you know popping

1:21 back up everywhere and so you know

1:24 networking became like the thing uh

1:27 Cisco which stock was kind of stagnating

1:29 for for for many years prior just

1:32 exploded and this is where Juniper

1:35 actually was also founded during that

1:36 same time to actually deliver on the

1:38 products to support all of this traffic

1:40 explosion that we were seeing and at the

1:43 time you know these data centers started

1:45 spurring everywhere and uh you know most

1:48 applications were quite simple client

1:50 server and so you had these

1:52 architectures which were more like north

1:54 south architecture if you recall the

1:56 access distribution core uh which were

1:59 kind of like the staple of the arch data

2:02 center architecture of the time kind of

2:04 led by Cisco uh fast forward a few years

2:08 uh around 2005 I would say uh we had

2:10 kind of a transition from you know what

2:13 were simpler services to more complex

2:16 Services more interesting offerings um

2:19 the start of the cloud right uh with

2:21 companies of course like Google but then

2:24 uh also Amazon and then actual public

2:27 Cloud which with AWS and uh and Azure uh

2:32 and so for these uh networks and for

2:35 these applications we know the the the

2:38 application itself the architecture of

2:40 the application changed it used to be

2:42 that you know you ask uh the server for

2:45 a request and then you just get a

2:46 response with these applications they

2:49 became more distributed and so for every

2:51 request that you had you know there was

2:52 a lot more chatter between the servers

2:55 and so we kind of moved from what were

2:57 north south application or the infr or

3:00 architectures to like these East West

3:02 architectures and so from the access

3:05 distribution core to like what what was

3:08 is is referred to as Leaf spine or Leaf

3:10 spine super spine that that support a

3:13 lot more East wwest traffic within the

3:16 data center at the time I was a at a

3:18 company that joined company early called

3:20 Arista uh and you know arguably they

3:23 kind of we led the way in kind of this

3:25 transition to this these Leaf spine uh

3:28 architectures and so that was in 2005

3:31 and with every one of those Evolutions

3:33 we see more explosion in traffic and in

3:36 networking traffic um then you know the

3:39 next moment that that is kind of seminal

3:42 is when of course the iPhone was

3:44 released and that's when everyone

3:46 realized that actually we can have apps

3:50 on these phones and then with these apps

3:52 anything that in the past you you had to

3:54 you have to go somewhere to do or to be

3:57 in front of your computer you actually

3:58 could do it right there from your phone

4:01 whether it's uh calling a taxi or uh

4:06 video conferencing with your U uh with

4:08 with your work uh or watching a movie or

4:12 transacting online buying something or

4:15 uh or doing a financial transaction

4:18 everything you you can think of you can

4:20 do now with your phone which means that

4:23 think about it for everything you do

4:25 you're touching a data center somewhere

4:27 you're touching a network somewhere

4:29 where so these networks not only were

4:31 growing massively in size but and they

4:34 were getting more complex but they were

4:36 becoming critical to everything we do

4:39 the networks are the foundation for

4:41 everything we do this is the phenomena

4:43 that is referred to as digital

4:46 transformation uh and uh the thing about

4:50 digital transformation is that it has

4:53 networking at at its core certainly with

4:56 the with the pandemic that everyone went

4:58 through this process has completely

5:00 accelerated and I would say Garner said

5:03 it best when they made this comment

5:04 which is that in order to be to succeed

5:07 a digital transformation you have to uh

5:10 transform your network first if you

5:13 don't then you're three times more

5:15 likely to fail at your digital

5:17 transformation and that kind that was

5:19 kind of the impetus for us actually

5:22 starting a company called abstra uh that

5:25 was back in

5:26 2014 and the goal was really to

5:29 transform the network to transform

5:32 network operations so that you no longer

5:35 have to manually uh operate your network

5:39 we wanted a network that really looks

5:41 and feels and is operated like a cloud

5:44 and so in order to do that we invented

5:46 this thing called intent based

5:48 networking the key there was that at the

5:51 end we wanted Network operators uh we

5:54 want to give them the ability not only

5:56 to move fast at the speed of the

5:58 business but the do it with utmost

6:00 reliability reliability was really the

6:03 key we wanted the network operator to

6:05 trust the software to run the network

6:09 better than they would themselves

6:10 actually in order of magnitude better

6:13 the same way in autopilot r or flies a

6:16 plane much more reliably than the pilot

6:19 themselves and so fast forward to today

6:23 uh we were acquired by Juniper we're

6:25 deployed uh across uh companies in more

6:28 than 7 countries making a difference

6:31 every day uh 83% reduction in Opex uh

6:35 multifold improvements in

6:38 reliability um essentially our customers

6:41 are getting the value of this Simplicity

6:43 the value of this reliability every

6:45 there these are a few of our customers

6:47 we have many many

6:49 more and so you would think okay so is

6:52 that it is there any more inflection

6:54 points and well you know uh over the

6:57 last few years and certainly the moment

6:59 that really captured everyone's

7:00 attention with Chad GP was Chad GPT back

7:03 in 2022 is the Advent of AI ML and yet

7:08 another inflection point yet another

7:11 explosion of traffic you know AI is of

7:14 course it captured the imagination of

7:17 everyone it's relevant to everyone we're

7:19 all using it today but what's critical

7:22 about Ai and what's really interesting

7:23 about AI is the kind of the what what it

7:27 what it take takes for the network and

7:29 for the infrastructure and for the data

7:31 centers to support all of the AI

7:33 applications that you know with all you

7:36 know all organizations want to use AI

7:39 for essentially you know there was this

7:42 blog from Andre which which I thought

7:44 was was really interesting where they

7:46 were kind of looking at where is the the

7:48 value chain for AI who's going to make

7:50 the most you know um you know like where

7:54 is the golden opportunity and uh at the

7:57 end of course AI is going to have a a

7:59 large value chain Lots there's going to

8:01 be lots of opportunity for everyone but

8:03 one thing that is 100% clear is that

8:07 it's going to at the end we're going to

8:08 need infrastructure infrastructure is

8:11 going to be you know the guaranteed

8:14 benefit of AI uh not just the gpus but

8:18 also the network we're going to have to

8:20 need to reinvent the network to support

8:23 these AI applications and if you look at

8:26 it then you know the data center um

8:29 traffic or the the number of data center

8:31 devices whichever way you want to

8:33 measure it essentially is just growing

8:35 and growing through every one of these

8:37 inflection points and you can see here

8:39 how AIML is kind of driving this new

8:42 traffic growth within Data Center and so

8:46 at the end you know the my message is

8:48 data center networking is thriving uh

8:51 it's continues to be relevant it

8:53 continues to be core and critical to

8:56 everything we do and of course with at

8:59 Juniper we're very interested in uh this

9:02 opportunity we' love to make a

9:04 difference in the life of our data

9:07 center customers and so uh then let me

9:10 tell you a bit about Juniper and and you

9:12 know what we do here when it comes to

9:14 the data center uh well first of all

9:18 we're very much relevant in the data

9:20 center market and we focus I would say

9:23 on large Enterprises Cloud customers and

9:28 uh more recently we're very focused on

9:31 the AI ml data center opportunity which

9:34 actually Cuts both across large

9:36 Enterprise and Cloud both Enterprises

9:39 and Cloud are building these AIML data

9:42 centers and so what products do we do we

9:46 do we uh do we deliver for those uh

9:48 customers you know at the from at the

9:50 basic level we have switches which are

9:53 kind of like the building blocks for

9:55 data center uh networks but then also we

9:57 have the operation software which at its

10:00 foundation leverages and utilizes abstra

10:04 both unram and U more and more in the

10:07 cloud but then you know net Juniper is

10:10 not just a data center company Juniper

10:13 while we're 100% focused on networking

10:15 that's all we do we have an entire

10:17 portfolio we actually have an endtoend

10:19 portfolio all the way from campus to

10:21 when and to security and so that's kind

10:24 of one of the benefits is that we can

10:26 leverage this extended portfolio when we

10:29 deliver data center solutions to our

10:30 customers adding on the the gateways the

10:34 security uh many virtual devices device

10:37 options and we'll get back to that but

10:39 essentially you know again with juniper

10:41 you know while we as in our business

10:43 units focused on data center Juniper has

10:45 the portfolio for in networking

10:48 Solutions end to end so you know going

10:51 talking about these devices we have a

10:53 whole portfolio uh you know we have

10:55 shallow buff shallow shallow buffer

10:58 device devices that are that are

11:01 relevant in for many applications

11:03 including IML applications for the

11:06 high-end we also have de buffer

11:08 applications we have Merchant silicon we

11:10 also have our own investment in our own

11:13 silicon and what one thing we're doing

11:15 is that we're boosting our switching for

11:18 portfolio for a IML we we actually

11:20 launched uh recently the tomahawk four

11:23 switches and this is kind of the start

11:25 of a many more products that we're going

11:28 to be delivering focusing on AI uh ML

11:32 and so that's kind of the basics now how

11:35 do we differentiate what is it that we

11:38 do differently than our competitors and

11:41 you know when I think of differentiators

11:42 the way I like I like to think about it

11:44 is what problems we're solving for our

11:47 customers uniquely and what I'd say is

11:50 that there are three pillars for our

11:52 differentiation one is operations and

11:54 this is something we're going to be

11:55 talking about a lot today second is

11:58 openness and third is endtoend Solutions

12:01 so let me talk about these starting with

12:05 operations so with operations as I said

12:09 abstra was the foundation and what we

12:10 did with abstra is this intent based

12:13 deterministic control essentially it's

12:15 software that runs your data center

12:18 Network day in and day out with the

12:21 speed to meet the business requirement

12:23 and with utmost reliability essentially

12:26 you want to trust your software to run

12:28 your network work that is kind of the

12:30 foundation for uh for uh our data center

12:35 operation strategy now in addition to

12:39 intent based and so that's that's intent

12:41 based so now in addition to intent based

12:42 what we're doing is that we're adding

12:45 other pillars to really kind of deliver

12:48 more value than anyone to really help

12:52 our Network operators operate their data

12:55 center with the Simplicity of the cloud

12:58 in fact with the utmost confidence and

13:01 so in addition to AI we're adding in in

13:05 addition to intent base we're adding in

13:07 Ai and so the way to think of it is that

13:11 intent base really gives you this

13:12 deterministic control you know there is

13:14 no room for probability when it comes to

13:18 running your data center the same way an

13:20 autopilot runs a plane when you tell it

13:22 to go to 10,000 ft you want it to go to

13:24 10,000 ft deterministically that's

13:27 essentially What in in based means but

13:30 while we're running the network with

13:33 this intent based approach we're

13:34 collecting tons of telemetry and this

13:37 Telemetry can be used and so what we do

13:39 is that we ship it to third-party

13:41 platforms but one of the platforms is

13:43 one that Juniper has which is Marvis we

13:46 have our own aiops layer which

13:49 essentially is a conversational

13:50 interface but also a data link and there

13:53 we can by sending all of the tra traffic

13:56 there and having the whole historical

13:58 perspective by applying AI you can now

14:00 deliver probabilistic insights and so

14:04 with intent based you get um tenfold

14:07 Improvement in reliability you get a uh

14:11 90% reduction in Opera operating

14:13 expenses but then when you also add AI

14:16 then what we have proven is that we can

14:19 also reduce the time to troubleshoot

14:22 problems reduce the number of tickets by

14:25 up to 90 to

14:26 95% and so by marrying this intent based

14:30 approach which is deterministic with the

14:32 probabilistic AIML approach essentially

14:36 where differentiating from everyone else

14:38 and adding that much more value to our

14:40 customers one one thing manour uh and

14:43 thank you it's great so far I love it uh

14:45 I would love to see as we get into

14:47 looking at it when we talk about

14:49 managing you know deterministic

14:51 probabilistic but also human driven the

14:54 important part here is like AI can tell

14:56 us what traffic is doing it can the app

14:59 awareness that all this stuff is great

15:01 but there's like business proprietary

15:03 you know sort of tribal knowledge that

15:05 sometimes we have I would love to see in

15:07 product kind of like how we can

15:09 influence determinism uh and sort of

15:12 like Drive the the basian inference with

15:15 a little bit of human

15:17 good you're you're right ahead of me all

15:20 right make sure this is awesome because

15:22 it's exactly what I was what I was going

15:23 to add right here and actually we're

15:25 going to show some of it today you know

15:27 the the third pillar and so like to

15:29 answer your question the third pillar

15:30 and you you talked about it the app

15:32 awareness we want to bring in the app

15:33 awareness at the end for

15:36 infrastructure uh the the the key asset

15:39 is the app right we what network

15:42 operators want to know is how is the

15:45 network performing for the app and we

15:47 want to be as Discerning as possible and

15:50 to your point some of these requirements

15:53 are very bespoke to the specific

15:56 customer and so what we need to do is

15:58 also have flexibility in how we kind of

16:02 put together these tests that leverage

16:04 all of this data so that we distill an

16:07 answer that a specific customer cares

16:10 about and so we're going to show some of

16:12 it today in terms of like the

16:13 flexibility of how you can build tests

16:16 but at the core of it is you need to

16:18 have the flow information and that's

16:19 something we just added today we added

16:22 recently we we launched it recently the

16:24 abstra flow data that's the most

16:27 contextual um uh most comprehensive flow

16:32 Telemetry uh of any data center

16:35 operations software on the market and

16:38 the reason again is because we want to

16:40 combine intent-based

16:41 Ai and application awareness together

16:44 and then have the ability to run tests

16:47 to be the most Discerning in terms of

16:49 whether or not the network is working

16:52 well for this specific app for this

16:54 specific use case okay and again we're

16:57 going to be showing uh more on on this

16:59 particular point today so that is

17:02 operations and this is very consistent

17:04 with Juniper's overarching strategy

17:06 experience first leading with operations

17:08 which is why they acquired abstra this

17:10 is why they acquired uh mist and that's

17:13 an area of core competency where we're

17:15 going to continue to invest the second

17:18 uh pillar of the Juniper data center

17:21 differentiation is openness

17:24 essentially the promise of networking

17:27 was that it is open

17:29 you the networking is so critical this

17:31 is why we have standards we don't want

17:32 to be locked into one specific vendor

17:35 this is why we've created things like

17:37 bgp and EPN VXL and so we at at Juniper

17:43 that's always been part of the DNA

17:45 because uniper kind of was founded as a

17:47 challenger to to the incumbent and

17:50 continues to be a challenger and at

17:52 abstra that was a core pillar we wanted

17:55 to deliver a solution that is

17:57 multivendor and so so although we were

17:59 acquired by Juniper we're continuing

18:02 down that path we support not only

18:04 Juniper but also uh switches from many

18:07 of uh of the other switch suppliers are

18:11 there in the market all of the

18:12 incumbents and the leading ones and so

18:16 uh if you deploy uh Juniper operations

18:19 then you can bring in uh not only

18:22 Juniper devices but Juniper devices from

18:25 the rest from from the rest of the the

18:28 the suppliers and so and when we say

18:30 that it's multivendor we don't just mean

18:33 that we support it but we also make sure

18:35 it works well so at the end we're kind

18:39 of giving you that same level of

18:40 reliability independently of your vendor

18:43 choice so that's one aspect of openness

18:46 which is unique to Juniper none of our

18:50 competitors has the other aspect to

18:52 openness is that you know at the end

18:54 we're part of an ecosystem you know this

18:57 this solution is is part of an ecosystem

18:59 and needs to work well in the context of

19:01 this ecosystem so we have open apis both

19:04 with our software with absr also with

19:06 all of the devices uh in that uh uh that

19:10 we deliver and we have Integrations with

19:14 platforms and solutions whether it's um

19:17 overlay Solutions or net devop Solutions

19:21 security Etc solutions that are

19:24 available on the market and one of the

19:27 tools that has become very popular to

19:29 manage Data Center and Cloud

19:32 environments is terraform and one of the

19:33 things we're going to Showcase today is

19:35 an integration with terraform as kind of

19:38 a unified tool to manage your data

19:40 center Network through Abra but also

19:43 your Club your your uh your public Cloud

19:46 so that's openness and the third is

19:48 endtoend Solutions and this is how kind

19:50 of it all comes together um we're kind

19:53 of taking as I mentioned Juniper is a

19:56 company that has not only focuses on

19:59 networking but has the entire portfolio

20:01 and so we want to really bring this to

20:03 bear so instead of selling products to

20:05 our customers what we want is to deliver

20:07 outcomes to deliver Solutions and so

20:10 whether you know customers think about

20:12 what use cases they have are they trying

20:13 to build an AI training cluster or a

20:16 VMware data center are they building an

20:19 eastwest three-stage Class A five-stage

20:21 class network is it an edge data center

20:25 whatever their uh their their use case

20:28 ISS what we want to do is deliver a

20:30 solution and when we what we mean with a

20:32 solution is uh a reference design that

20:35 includes the devices themselves the

20:37 topology how they're interconnected what

20:40 protocols they use what software manages

20:42 them of course primarily at its core

20:45 abstra but not only that we also are

20:47 bringing in the routers the gateways the

20:50 firewalls as part of the design um you

20:53 can use juniper uh devices but also you

20:55 can bring in devices from the

20:56 competition and then what we want to do

20:58 is give it to you in a way that is

21:01 tested end to end so you know it's going

21:04 to work and so we're kind of going to be

21:06 leading with this validated designs

21:09 because you know while some customers

21:11 are interested in kind of tailoring

21:13 their Solutions having some bespoke

21:14 designs most customers want solutions

21:18 that just work end to endend and that is

21:21 what at Juniper we want to deliver for

21:23 them and in fact one of the things we're

21:24 going to show today is how we can EAS

21:28 easily deploy build and operate an AI

21:32 cluster using Abra essentially

21:35 delivering on this validated designed

21:37 for an AI cluster and

21:41 so what did you mean for taish

21:44 troubleshooting or brother what is the

21:46 context of troubleshooting which is uh

21:50 when the AI and ml considering a trouble

21:54 when I'm shooting the trouble so uh if

21:57 is something Proactive or deterministic

22:00 as as it told and uh what if I already

22:06 know that the the the network will be uh

22:08 busy because I have a streaming so I

22:12 know that it's busy but the the

22:15 troubleshooting

22:17 um made up of abstract could consider

22:20 this an anomaly but uh I've been um um

22:25 so when we say when typically with

22:27 troubleshooting troubleshooting an issue

22:29 with an application like generally your

22:31 application is not working well right

22:33 and okay you know usually when it's when

22:36 it when it's when you're part of the

22:38 networking team you're trying to figure

22:40 out whether or not you know something

22:42 that's under your control is at fault

22:44 right whether the network is at fault

22:47 and you know if with with with abstra if

22:50 you have all of the knowledge of all of

22:52 the pieces then you can

22:53 deterministically say Hey you know there

22:54 is a problem here or there but uh in in

22:58 many cases you don't have all of that

23:00 knowledge especially when you're you're

23:02 uh you're looking at a multi- domain

23:05 infrastructure when you're kind of

23:07 looking at a problem and you're not sure

23:08 whether it's really in the data center

23:10 or whether it's somewhere on the on the

23:12 W side uh or in fact you're not even

23:16 sure whether it's on the you know it's

23:18 it's within the network infrastructure

23:20 itself and so when you have the

23:22 knowledge you can use a deterministic

23:24 approach but if you don't have all of

23:26 the knowledge then by looking at your

23:28 historical data and applying AI models

23:32 to your uh your uh your your data leg

23:35 all of the Telemetry that you have

23:37 throughout the history then you know

23:40 these AI models have the ability to kind

23:42 of point you in the right direction

23:44 right essentially you look the at all of

23:47 the knowledge that you have all of over

23:49 the over over time and you have an

23:52 ability to go and say okay well

23:53 probabilistically I believe that you you

23:55 know it's the problem is likely to be

23:58 around there right and so it kind of is

24:00 help helps you guide you it's not

24:02 guaranteed it's not deterministic but it

24:05 guides you right okay that's clear thank

24:07 you yeah so it's driven by uh Ai and ml

24:12 so that if it will prepare dynamically

24:16 some kind of templates to say this is

24:18 wrong or this is acting correctly and

24:21 and so on is it right yes correct okay

24:24 thank you thank

24:26 you

0:09 um you may know me as more most recently

0:12 as the uh founder of Abra uh but given

0:16 that recently I got an an expanded role

0:18 focusing on the data center all of the

0:20 data center products for Juniper what I

0:22 thought is talk more about the data

0:25 center and uh you know the the all of

0:28 the things we're bringing as in Juniper

0:30 to the table when it comes to uh really

0:33 solving problems for our data center uh

0:37 customers and so that's going to be the

0:39 focus today and uh we're going to also

0:41 showcase a bunch of those capabilities

0:44 uh you know throughout the session and

0:46 so but before I jump in uh what I

0:49 thought I'd do is kind of step back and

0:51 talk about the evolution of the data

0:54 center and coincidentally I'm starting

0:56 in September of 2 of 1995 which is when

0:59 actually I came to the US uh to attend

1:02 Stanford and um that was coincidentally

1:05 the same uh month where Netscape

1:08 actually went IPO if you recall uh at

1:11 the time if you were there uh it was a

1:14 crazy time where every company was kind

1:16 of rushing to go online um you know

1:19 websites were U Were you know popping

1:21 back up everywhere and so you know

1:24 networking became like the thing uh

1:27 Cisco which stock was kind of stagnating

1:29 for for for many years prior just

1:32 exploded and this is where Juniper

1:35 actually was also founded during that

1:36 same time to actually deliver on the

1:38 products to support all of this traffic

1:40 explosion that we were seeing and at the

1:43 time you know these data centers started

1:45 spurring everywhere and uh you know most

1:48 applications were quite simple client

1:50 server and so you had these

1:52 architectures which were more like north

1:54 south architecture if you recall the

1:56 access distribution core uh which were

1:59 kind of like the staple of the arch data

2:02 center architecture of the time kind of

2:04 led by Cisco uh fast forward a few years

2:08 uh around 2005 I would say uh we had

2:10 kind of a transition from you know what

2:13 were simpler services to more complex

2:16 Services more interesting offerings um

2:19 the start of the cloud right uh with

2:21 companies of course like Google but then

2:24 uh also Amazon and then actual public

2:27 Cloud which with AWS and uh and Azure uh

2:32 and so for these uh networks and for

2:35 these applications we know the the the

2:38 application itself the architecture of

2:40 the application changed it used to be

2:42 that you know you ask uh the server for

2:45 a request and then you just get a

2:46 response with these applications they

2:49 became more distributed and so for every

2:51 request that you had you know there was

2:52 a lot more chatter between the servers

2:55 and so we kind of moved from what were

2:57 north south application or the infr or

3:00 architectures to like these East West

3:02 architectures and so from the access

3:05 distribution core to like what what was

3:08 is is referred to as Leaf spine or Leaf

3:10 spine super spine that that support a

3:13 lot more East wwest traffic within the

3:16 data center at the time I was a at a

3:18 company that joined company early called

3:20 Arista uh and you know arguably they

3:23 kind of we led the way in kind of this

3:25 transition to this these Leaf spine uh

3:28 architectures and so that was in 2005

3:31 and with every one of those Evolutions

3:33 we see more explosion in traffic and in

3:36 networking traffic um then you know the

3:39 next moment that that is kind of seminal

3:42 is when of course the iPhone was

3:44 released and that's when everyone

3:46 realized that actually we can have apps

3:50 on these phones and then with these apps

3:52 anything that in the past you you had to

3:54 you have to go somewhere to do or to be

3:57 in front of your computer you actually

3:58 could do it right there from your phone

4:01 whether it's uh calling a taxi or uh

4:06 video conferencing with your U uh with

4:08 with your work uh or watching a movie or

4:12 transacting online buying something or

4:15 uh or doing a financial transaction

4:18 everything you you can think of you can

4:20 do now with your phone which means that

4:23 think about it for everything you do

4:25 you're touching a data center somewhere

4:27 you're touching a network somewhere

4:29 where so these networks not only were

4:31 growing massively in size but and they

4:34 were getting more complex but they were

4:36 becoming critical to everything we do

4:39 the networks are the foundation for

4:41 everything we do this is the phenomena

4:43 that is referred to as digital

4:46 transformation uh and uh the thing about

4:50 digital transformation is that it has

4:53 networking at at its core certainly with

4:56 the with the pandemic that everyone went

4:58 through this process has completely

5:00 accelerated and I would say Garner said

5:03 it best when they made this comment

5:04 which is that in order to be to succeed

5:07 a digital transformation you have to uh

5:10 transform your network first if you

5:13 don't then you're three times more

5:15 likely to fail at your digital

5:17 transformation and that kind that was

5:19 kind of the impetus for us actually

5:22 starting a company called abstra uh that

5:25 was back in

5:26 2014 and the goal was really to

5:29 transform the network to transform

5:32 network operations so that you no longer

5:35 have to manually uh operate your network

5:39 we wanted a network that really looks

5:41 and feels and is operated like a cloud

5:44 and so in order to do that we invented

5:46 this thing called intent based

5:48 networking the key there was that at the

5:51 end we wanted Network operators uh we

5:54 want to give them the ability not only

5:56 to move fast at the speed of the

5:58 business but the do it with utmost

6:00 reliability reliability was really the

6:03 key we wanted the network operator to

6:05 trust the software to run the network

6:09 better than they would themselves

6:10 actually in order of magnitude better

6:13 the same way in autopilot r or flies a

6:16 plane much more reliably than the pilot

6:19 themselves and so fast forward to today

6:23 uh we were acquired by Juniper we're

6:25 deployed uh across uh companies in more

6:28 than 7 countries making a difference

6:31 every day uh 83% reduction in Opex uh

6:35 multifold improvements in

6:38 reliability um essentially our customers

6:41 are getting the value of this Simplicity

6:43 the value of this reliability every

6:45 there these are a few of our customers

6:47 we have many many

6:49 more and so you would think okay so is

6:52 that it is there any more inflection

6:54 points and well you know uh over the

6:57 last few years and certainly the moment

6:59 that really captured everyone's

7:00 attention with Chad GP was Chad GPT back

7:03 in 2022 is the Advent of AI ML and yet

7:08 another inflection point yet another

7:11 explosion of traffic you know AI is of

7:14 course it captured the imagination of

7:17 everyone it's relevant to everyone we're

7:19 all using it today but what's critical

7:22 about Ai and what's really interesting

7:23 about AI is the kind of the what what it

7:27 what it take takes for the network and

7:29 for the infrastructure and for the data

7:31 centers to support all of the AI

7:33 applications that you know with all you

7:36 know all organizations want to use AI

7:39 for essentially you know there was this

7:42 blog from Andre which which I thought

7:44 was was really interesting where they

7:46 were kind of looking at where is the the

7:48 value chain for AI who's going to make

7:50 the most you know um you know like where

7:54 is the golden opportunity and uh at the

7:57 end of course AI is going to have a a

7:59 large value chain Lots there's going to

8:01 be lots of opportunity for everyone but

8:03 one thing that is 100% clear is that

8:07 it's going to at the end we're going to

8:08 need infrastructure infrastructure is

8:11 going to be you know the guaranteed

8:14 benefit of AI uh not just the gpus but

8:18 also the network we're going to have to

8:20 need to reinvent the network to support

8:23 these AI applications and if you look at

8:26 it then you know the data center um

8:29 traffic or the the number of data center

8:31 devices whichever way you want to

8:33 measure it essentially is just growing

8:35 and growing through every one of these

8:37 inflection points and you can see here

8:39 how AIML is kind of driving this new

8:42 traffic growth within Data Center and so

8:46 at the end you know the my message is

8:48 data center networking is thriving uh

8:51 it's continues to be relevant it

8:53 continues to be core and critical to

8:56 everything we do and of course with at

8:59 Juniper we're very interested in uh this

9:02 opportunity we' love to make a

9:04 difference in the life of our data

9:07 center customers and so uh then let me

9:10 tell you a bit about Juniper and and you

9:12 know what we do here when it comes to

9:14 the data center uh well first of all

9:18 we're very much relevant in the data

9:20 center market and we focus I would say

9:23 on large Enterprises Cloud customers and

9:28 uh more recently we're very focused on

9:31 the AI ml data center opportunity which

9:34 actually Cuts both across large

9:36 Enterprise and Cloud both Enterprises

9:39 and Cloud are building these AIML data

9:42 centers and so what products do we do we

9:46 do we uh do we deliver for those uh

9:48 customers you know at the from at the

9:50 basic level we have switches which are

9:53 kind of like the building blocks for

9:55 data center uh networks but then also we

9:57 have the operation software which at its

10:00 foundation leverages and utilizes abstra

10:04 both unram and U more and more in the

10:07 cloud but then you know net Juniper is

10:10 not just a data center company Juniper

10:13 while we're 100% focused on networking

10:15 that's all we do we have an entire

10:17 portfolio we actually have an endtoend

10:19 portfolio all the way from campus to

10:21 when and to security and so that's kind

10:24 of one of the benefits is that we can

10:26 leverage this extended portfolio when we

10:29 deliver data center solutions to our

10:30 customers adding on the the gateways the

10:34 security uh many virtual devices device

10:37 options and we'll get back to that but

10:39 essentially you know again with juniper

10:41 you know while we as in our business

10:43 units focused on data center Juniper has

10:45 the portfolio for in networking

10:48 Solutions end to end so you know going

10:51 talking about these devices we have a

10:53 whole portfolio uh you know we have

10:55 shallow buff shallow shallow buffer

10:58 device devices that are that are

11:01 relevant in for many applications

11:03 including IML applications for the

11:06 high-end we also have de buffer

11:08 applications we have Merchant silicon we

11:10 also have our own investment in our own

11:13 silicon and what one thing we're doing

11:15 is that we're boosting our switching for

11:18 portfolio for a IML we we actually

11:20 launched uh recently the tomahawk four

11:23 switches and this is kind of the start

11:25 of a many more products that we're going

11:28 to be delivering focusing on AI uh ML

11:32 and so that's kind of the basics now how

11:35 do we differentiate what is it that we

11:38 do differently than our competitors and

11:41 you know when I think of differentiators

11:42 the way I like I like to think about it

11:44 is what problems we're solving for our

11:47 customers uniquely and what I'd say is

11:50 that there are three pillars for our

11:52 differentiation one is operations and

11:54 this is something we're going to be

11:55 talking about a lot today second is

11:58 openness and third is endtoend Solutions

12:01 so let me talk about these starting with

12:05 operations so with operations as I said

12:09 abstra was the foundation and what we

12:10 did with abstra is this intent based

12:13 deterministic control essentially it's

12:15 software that runs your data center

12:18 Network day in and day out with the

12:21 speed to meet the business requirement

12:23 and with utmost reliability essentially

12:26 you want to trust your software to run

12:28 your network work that is kind of the

12:30 foundation for uh for uh our data center

12:35 operation strategy now in addition to

12:39 intent based and so that's that's intent

12:41 based so now in addition to intent based

12:42 what we're doing is that we're adding

12:45 other pillars to really kind of deliver

12:48 more value than anyone to really help

12:52 our Network operators operate their data

12:55 center with the Simplicity of the cloud

12:58 in fact with the utmost confidence and

13:01 so in addition to AI we're adding in in

13:05 addition to intent base we're adding in

13:07 Ai and so the way to think of it is that

13:11 intent base really gives you this

13:12 deterministic control you know there is

13:14 no room for probability when it comes to

13:18 running your data center the same way an

13:20 autopilot runs a plane when you tell it

13:22 to go to 10,000 ft you want it to go to

13:24 10,000 ft deterministically that's

13:27 essentially What in in based means but

13:30 while we're running the network with

13:33 this intent based approach we're

13:34 collecting tons of telemetry and this

13:37 Telemetry can be used and so what we do

13:39 is that we ship it to third-party

13:41 platforms but one of the platforms is

13:43 one that Juniper has which is Marvis we

13:46 have our own aiops layer which

13:49 essentially is a conversational

13:50 interface but also a data link and there

13:53 we can by sending all of the tra traffic

13:56 there and having the whole historical

13:58 perspective by applying AI you can now

14:00 deliver probabilistic insights and so

14:04 with intent based you get um tenfold

14:07 Improvement in reliability you get a uh

14:11 90% reduction in Opera operating

14:13 expenses but then when you also add AI

14:16 then what we have proven is that we can

14:19 also reduce the time to troubleshoot

14:22 problems reduce the number of tickets by

14:25 up to 90 to

14:26 95% and so by marrying this intent based

14:30 approach which is deterministic with the

14:32 probabilistic AIML approach essentially

14:36 where differentiating from everyone else

14:38 and adding that much more value to our

14:40 customers one one thing manour uh and

14:43 thank you it's great so far I love it uh

14:45 I would love to see as we get into

14:47 looking at it when we talk about

14:49 managing you know deterministic

14:51 probabilistic but also human driven the

14:54 important part here is like AI can tell

14:56 us what traffic is doing it can the app

14:59 awareness that all this stuff is great

15:01 but there's like business proprietary

15:03 you know sort of tribal knowledge that

15:05 sometimes we have I would love to see in

15:07 product kind of like how we can

15:09 influence determinism uh and sort of

15:12 like Drive the the basian inference with

15:15 a little bit of human

15:17 good you're you're right ahead of me all

15:20 right make sure this is awesome because

15:22 it's exactly what I was what I was going

15:23 to add right here and actually we're

15:25 going to show some of it today you know

15:27 the the third pillar and so like to

15:29 answer your question the third pillar

15:30 and you you talked about it the app

15:32 awareness we want to bring in the app

15:33 awareness at the end for

15:36 infrastructure uh the the the key asset

15:39 is the app right we what network

15:42 operators want to know is how is the

15:45 network performing for the app and we

15:47 want to be as Discerning as possible and

15:50 to your point some of these requirements

15:53 are very bespoke to the specific

15:56 customer and so what we need to do is

15:58 also have flexibility in how we kind of

16:02 put together these tests that leverage

16:04 all of this data so that we distill an

16:07 answer that a specific customer cares

16:10 about and so we're going to show some of

16:12 it today in terms of like the

16:13 flexibility of how you can build tests

16:16 but at the core of it is you need to

16:18 have the flow information and that's

16:19 something we just added today we added

16:22 recently we we launched it recently the

16:24 abstra flow data that's the most

16:27 contextual um uh most comprehensive flow

16:32 Telemetry uh of any data center

16:35 operations software on the market and

16:38 the reason again is because we want to

16:40 combine intent-based

16:41 Ai and application awareness together

16:44 and then have the ability to run tests

16:47 to be the most Discerning in terms of

16:49 whether or not the network is working

16:52 well for this specific app for this

16:54 specific use case okay and again we're

16:57 going to be showing uh more on on this

16:59 particular point today so that is

17:02 operations and this is very consistent

17:04 with Juniper's overarching strategy

17:06 experience first leading with operations

17:08 which is why they acquired abstra this

17:10 is why they acquired uh mist and that's

17:13 an area of core competency where we're

17:15 going to continue to invest the second

17:18 uh pillar of the Juniper data center

17:21 differentiation is openness

17:24 essentially the promise of networking

17:27 was that it is open

17:29 you the networking is so critical this

17:31 is why we have standards we don't want

17:32 to be locked into one specific vendor

17:35 this is why we've created things like

17:37 bgp and EPN VXL and so we at at Juniper

17:43 that's always been part of the DNA

17:45 because uniper kind of was founded as a

17:47 challenger to to the incumbent and

17:50 continues to be a challenger and at

17:52 abstra that was a core pillar we wanted

17:55 to deliver a solution that is

17:57 multivendor and so so although we were

17:59 acquired by Juniper we're continuing

18:02 down that path we support not only

18:04 Juniper but also uh switches from many

18:07 of uh of the other switch suppliers are

18:11 there in the market all of the

18:12 incumbents and the leading ones and so

18:16 uh if you deploy uh Juniper operations

18:19 then you can bring in uh not only

18:22 Juniper devices but Juniper devices from

18:25 the rest from from the rest of the the

18:28 the suppliers and so and when we say

18:30 that it's multivendor we don't just mean

18:33 that we support it but we also make sure

18:35 it works well so at the end we're kind

18:39 of giving you that same level of

18:40 reliability independently of your vendor

18:43 choice so that's one aspect of openness

18:46 which is unique to Juniper none of our

18:50 competitors has the other aspect to

18:52 openness is that you know at the end

18:54 we're part of an ecosystem you know this

18:57 this solution is is part of an ecosystem

18:59 and needs to work well in the context of

19:01 this ecosystem so we have open apis both

19:04 with our software with absr also with

19:06 all of the devices uh in that uh uh that

19:10 we deliver and we have Integrations with

19:14 platforms and solutions whether it's um

19:17 overlay Solutions or net devop Solutions

19:21 security Etc solutions that are

19:24 available on the market and one of the

19:27 tools that has become very popular to

19:29 manage Data Center and Cloud

19:32 environments is terraform and one of the

19:33 things we're going to Showcase today is

19:35 an integration with terraform as kind of

19:38 a unified tool to manage your data

19:40 center Network through Abra but also

19:43 your Club your your uh your public Cloud

19:46 so that's openness and the third is

19:48 endtoend Solutions and this is how kind

19:50 of it all comes together um we're kind

19:53 of taking as I mentioned Juniper is a

19:56 company that has not only focuses on

19:59 networking but has the entire portfolio

20:01 and so we want to really bring this to

20:03 bear so instead of selling products to

20:05 our customers what we want is to deliver

20:07 outcomes to deliver Solutions and so

20:10 whether you know customers think about

20:12 what use cases they have are they trying

20:13 to build an AI training cluster or a

20:16 VMware data center are they building an

20:19 eastwest three-stage Class A five-stage

20:21 class network is it an edge data center

20:25 whatever their uh their their use case

20:28 ISS what we want to do is deliver a

20:30 solution and when we what we mean with a

20:32 solution is uh a reference design that

20:35 includes the devices themselves the

20:37 topology how they're interconnected what

20:40 protocols they use what software manages

20:42 them of course primarily at its core

20:45 abstra but not only that we also are

20:47 bringing in the routers the gateways the

20:50 firewalls as part of the design um you

20:53 can use juniper uh devices but also you

20:55 can bring in devices from the

20:56 competition and then what we want to do

20:58 is give it to you in a way that is

21:01 tested end to end so you know it's going

21:04 to work and so we're kind of going to be

21:06 leading with this validated designs

21:09 because you know while some customers

21:11 are interested in kind of tailoring

21:13 their Solutions having some bespoke

21:14 designs most customers want solutions

21:18 that just work end to endend and that is

21:21 what at Juniper we want to deliver for

21:23 them and in fact one of the things we're

21:24 going to show today is how we can EAS

21:28 easily deploy build and operate an AI

21:32 cluster using Abra essentially

21:35 delivering on this validated designed

21:37 for an AI cluster and

21:41 so what did you mean for taish

21:44 troubleshooting or brother what is the

21:46 context of troubleshooting which is uh

21:50 when the AI and ml considering a trouble

21:54 when I'm shooting the trouble so uh if

21:57 is something Proactive or deterministic

22:00 as as it told and uh what if I already

22:06 know that the the the network will be uh

22:08 busy because I have a streaming so I

22:12 know that it's busy but the the

22:15 troubleshooting

22:17 um made up of abstract could consider

22:20 this an anomaly but uh I've been um um

22:25 so when we say when typically with

22:27 troubleshooting troubleshooting an issue

22:29 with an application like generally your

22:31 application is not working well right

22:33 and okay you know usually when it's when

22:36 it when it's when you're part of the

22:38 networking team you're trying to figure

22:40 out whether or not you know something

22:42 that's under your control is at fault

22:44 right whether the network is at fault

22:47 and you know if with with with abstra if

22:50 you have all of the knowledge of all of

22:52 the pieces then you can

22:53 deterministically say Hey you know there

22:54 is a problem here or there but uh in in

22:58 many cases you don't have all of that

23:00 knowledge especially when you're you're

23:02 uh you're looking at a multi- domain

23:05 infrastructure when you're kind of

23:07 looking at a problem and you're not sure

23:08 whether it's really in the data center

23:10 or whether it's somewhere on the on the

23:12 W side uh or in fact you're not even

23:16 sure whether it's on the you know it's

23:18 it's within the network infrastructure

23:20 itself and so when you have the

23:22 knowledge you can use a deterministic

23:24 approach but if you don't have all of

23:26 the knowledge then by looking at your

23:28 historical data and applying AI models

23:32 to your uh your uh your your data leg

23:35 all of the Telemetry that you have

23:37 throughout the history then you know

23:40 these AI models have the ability to kind

23:42 of point you in the right direction

23:44 right essentially you look the at all of

23:47 the knowledge that you have all of over

23:49 the over over time and you have an

23:52 ability to go and say okay well

23:53 probabilistically I believe that you you

23:55 know it's the problem is likely to be

23:58 around there right and so it kind of is

24:00 help helps you guide you it's not

24:02 guaranteed it's not deterministic but it

24:05 guides you right okay that's clear thank

24:07 you yeah so it's driven by uh Ai and ml

24:12 so that if it will prepare dynamically

24:16 some kind of templates to say this is

24:18 wrong or this is acting correctly and

24:21 and so on is it right yes correct okay

24:24 thank you thank

24:26 you