GPYOU: Building and Operating your AI Infrastructure with Juniper Networks

0:00 so my name is Kyle Baxter i head up the

0:03 Abstra product team today we're going to

0:06 be talking about GPU on how you are

0:10 going to be able to manage and deploy

0:13 your AI

0:15 infrastructure so we're going to walk

0:17 you through your AI journey we're going

0:20 to give you a quick intro on what is

0:23 Abstra for those that don't know and

0:26 then we'll get into how you use Abstra

0:29 to design your AI data center how you

0:35 manage your data center at scale for

0:38 deployment and how you operate your data

0:43 center and so let's get into a little

0:46 intro on what is

0:50 Abstra so I think everybody will agree

0:53 here that you can't build a great

0:56 building without a great foundation so

0:58 if you have a weak foundation you're

1:00 going to have problems in the network is

1:03 the foundation in AI and ML training and

1:06 inference jobs the problem is the

1:10 network is the problem there's always

1:13 lack of insight how do you know why

1:15 you're training jobs slow why are you

1:17 getting

1:18 congestion how do you deploy faster when

1:21 there's business needs to go deploy you

1:24 know new fabric or expand faster how do

1:26 you get that speed and get that reliably

1:28 um and then how do you prevent outages

1:32 and make sure your network is operating

1:35 reliable this is is where Juniper comes

1:38 in um because the network is complex

1:42 there's so many elements that make up a

1:45 network and when you look at you know

1:47 spines leaves and all the links and

1:49 we're amplituding and amplifying that

1:52 with AI when we're talking about 800 gig

1:55 networks and in the future 1.6 and

1:57 beyond that are going to get some crazy

1:59 speeds that if anything goes wrong it's

2:01 going to go wrong fast and the problem

2:04 with most network management tools out

2:07 there is they flood you with data which

2:10 is nice but then you're looking for the

2:12 needle in the haststack you're just

2:13 looking through all these data points

2:15 trying to figure out what means what and

2:18 that's what we've done differently at

2:21 Juniper we've brought together our

2:24 technology together to build something

2:26 different and something better and so we

2:30 brought Abstra which brings the

2:33 industry's only contextual graph

2:35 database with intentbased networking and

2:38 designs with a vendor agnostic approach

2:41 we've combined that with Mist in the AI

2:44 there to bring with the industry's only

2:46 AI native platform and Marvis virtual

2:49 network assistant to the data center and

2:52 so we want to look at the data center a

2:54 little differently rather than as a

2:56 collection of individual switches that

2:57 you're configuring one by one we want to

2:59 look at it holistically as a solution on

3:03 how we can deliver the right

3:06 outcomes and we do that by cutting

3:09 through the complexity instead of seeing

3:13 all these random events and dots and

3:15 trying to figure out what's going on

3:16 Abstra cuts through that complexity by

3:19 bringing context to that data we can

3:22 pinpoint exactly what's the root cause

3:25 that you need to address and fix but

3:27 what is what are related symptoms that

3:29 you can ignore that you don't need to

3:32 worry about that you can shift all that

3:33 aside and then what are the impacted

3:36 applications or training jobs because of

3:39 that issue that we have pinpointed we

3:41 can do all of that for you and we'll

3:43 walk through a lot of that today and

3:47 what this delivers is a complete

3:49 solution from design deploy assure that

3:54 can run on our industry-leading

3:57 um switches and switching portfolio with

4:00 the integrated solution from security

4:02 that we talked about in another session

4:05 and be able to bring that to you

4:10 and we're going to focus a lot about on

4:13 the training side of AI and ML clusters

4:16 today but I want to make sure it's clear

4:19 that Aster can manage more than that

4:21 we've been managing data centers with

4:24 all over the g globe for some of the

4:26 largest enterprises out there that we

4:29 can do not only the backend networks

4:30 that we'll talk about today but storage

4:32 networks inference networks or any kind

4:34 of other data center network we have the

4:37 designs we had the flexibility to manage

4:39 any design and you can use one tool and

4:42 we'll see that here in uh in a later

4:44 session we talk about operating on how

4:47 we can use one tool to manage multiple

4:50 different networks