Mobile World Live Interview with Raj Yavatkar about AI Native Networks

0:03 I'm here at mwc Barcelona with the CTO

0:06 of Juniper Networks Raj yatka thank you

0:09 very much for joining us today thank you

0:11 Chris why are AI based networks

0:14 necessary and what Innovations are

0:16 coming to life because of this

0:17 technology so if you look at

0:19 traditionally the networks have always

0:21 been operated manually right so are

0:24 there two end consumers of the network

0:27 end users which have devices like I

0:30 phones and all and the end to operators

0:32 who operate the networks operators

0:35 always have been working with multiple

0:36 dashboards doing things manually in the

0:39 world of AI we had to move away from

0:41 that where they should be just trusting

0:44 the network to work Network should be

0:47 invisible if you look at end users in

0:49 the iPhone generation people have got

0:52 used to uh Network that just works on

0:56 our iPhone we don't care about it so our

0:59 idea behind AI for networking is to

1:01 build this what we call Modern AI native

1:04 networking platform based on AI Ops

1:07 which makes Network Invisible by

1:09 collecting right s of telemetry and

1:12 using machine learning to make sure

1:14 Network stays operational so one of the

1:16 things we have done is that last N9

1:18 years we have been working on collecting

1:19 data from the network applying data

1:21 science for n and insights so that based

1:25 on Security based on realtime responses

1:29 and network that that's reliable we can

1:31 provide connectivity using what we call

1:33 AI native networking platform it has

1:35 really two aspects one is apply AI for

1:38 networking so operations become

1:40 simplified and automatic and the other

1:42 is networking for AI so you can build

1:44 networks to support

1:45 AI so you've enhancing the power of AI

1:48 to make connections better but there's

1:49 another side to this as well about AI

1:51 workloads can you tell us a little bit

1:53 more about that that that's a very good

1:55 point so if you look at um generative AI

1:58 the latest Revolution it's all about

2:00 training large language models and when

2:03 you train these models you use clusters

2:05 of gpus and this is the most expensive

2:09 resource when you do that right and you

2:11 cannot you have to keep them occupied

2:13 you cannot leave them stranded waiting

2:15 for response from other gpus then the

2:17 network plays a very important role to

2:19 provide very low low latency High

2:22 throughput reliable communication that's

2:24 a network fabric for training AI

2:27 clusters so what we're doing is that we

2:30 have abstar based Network automation it

2:33 really helps with two aspects of the

2:34 network Day Zero where it's intent based

2:37 you express the intent and it

2:39 automatically configures and Designs the

2:40 network then at day two operational time

2:43 it continuously monitors the status of

2:45 the network and checks it against the

2:47 intent if it deviates it automatically

2:49 corrects it so that sort of closed loop

2:52 automation helps to create a network

2:54 fabric that's right for the AI training

2:56 models but on top of that what we do is

2:59 we bring our for a Marvis AI engine

3:01 which has a conversational interface but

3:03 also it has a machine learning

3:04 capabilities to take the output from

3:06 abstra events alerts and apply machine

3:09 learning for automated Diagnostics and

3:12 correction to the network so those two

3:14 aspects make really the ability to do

3:16 what we call networking for AI so one of

3:19 the big discussion topics at the moment

3:21 is the impact of AI on sustainability

3:23 efforts how is Juniper leading in this

3:25 area so as we look at the AI workloads

3:27 take lots of computing and networking

3:29 power it's important to be sustainable

3:31 so as part of the Juniper Beyond Labs

3:33 initiative we are trying to make sure

3:35 across all of our product development we

3:37 apply sustainable uh practices including

3:40 new designs so apply machine learning

3:42 for example to reduce the amount of

3:44 power used by the network by observing

3:47 where the hotpots in the network and

3:49 which network parts are not being used

3:51 and using machine learning to start

3:53 predicting usage and optimized Network

3:56 we're also doing that across

3:57 decarbonization and circularity to

4:00 reduce the amount of energy we use as in

4:02 our operations but also how we recycle

4:05 the parts or components we use in our

4:07 Hardware platforms thank you very much

4:09 for your insights today thank you

4:12 [Music]

4:19 Chris