Your Private AI Data Center, as Easy as Cloud with Juniper Networks

0:10 so yeah good morning good afternoon good evening wherever in the world you are um my name is Nick Dave I'm a director of product management and I'm here to talk

0:16 to you a little bit about how we can build all of these things that my colleagues just told you

0:22 about um because like you heard from preul and maner the workloads the applications the use cases that we're

0:29 trying to drive out of modern data centers are pushing scale to ridiculous

0:34 levels um when you look at those designs on the uh on the the previous presenter

0:41 slides you'll see complexity and you'll see uh connectivity like we've never had to provision before in data centers and

0:48 as a former network engineer I do not want to be doing any of this by hand um so I'm here to talk to you today about

0:55 how you can build all of this and not be overly bothered with the details

1:00 uh of course our answer for this is taking all of that science all of that research that we've come up with in our

1:06 Labs all of the testing and qualification that we do around our products and our use cases and we pack

1:11 those into a set of validated designs uh the jvd I'll be talking about this

1:17 morning is our AI jvd which talks about how to build uh a readymade AI data

1:23 center um in addition to the the types of things you'd normally expect in a design white paper right things like

1:29 architecture diagrams and configuration protocol explanations and a little bit of chitter chatter about the

1:35 encapsulations uh we actually include one more important thing and that's the automation actually required to make

1:41 these things spring out of the box and come to life and that was a question earlier I think about like how do we

1:46 actually package a jvd and what do you get right you don't just download a PDF I'll show you later and Raj will

1:53 actually show you for real how we take automation that we design around our jbds and instantiate real data centers

1:59 using the power of abstra uh we try and bake in all of the learning that we do in our Labs into

2:06 these jvds actually we posted the AI jbd this morning I think if you go to juniper.net you'll see that updated

2:11 document uh and so we try to keep these things uh up to date and packed full of our latest guidance on how to deploy the

2:18 most complex networks um of course even for AI there's no one- siiz fits-all

2:23 right and so part of the automation that we build actually lets us T-shirt size these things appropriately to your use

2:29 cas case uh we'll talk in Practical terms about where that can be handy like when we're sharing a big AI lab that we

2:35 all want for our very own um but one of the Practical things we talked about this morning was over subscription

2:41 ratios right or even the dimensions or the the physical Hardware making up one

2:46 of these AI data centers all of that gets packed in as parameters into our automation so that you can tweak and

2:51 tune your AI jvd to match your network requirement Central to all of this is AB

3:00 uh and there's a very good reason for that as well right because like when manur said that folks are bringing back workloads from the public cloud and and

3:07 starting to reestablish uh large private cloud data centers the one thing that he forgot to mention it's very important

3:13 that people brought back from the public Cloud it's expectations uh people are very used to

3:19 uh and and operators application owners are very used to clicking a button and receiving the thing that they asked for

3:25 they don't go into a protracted set of meetings we don't open tickets we don't have discussion about it we press a

3:31 button and sometimes we slide our credit card but usually we just get the thing that we wanted so that's the experience

3:37 that we have to beat and I'm happy to say that all of the technologies that make that experience possible in the

3:43 public Cloud are possible in your own on Prem DCS as well uh abstra gives you that Central

3:50 Point to control an entire data center so you can look at that as offering up a cloudlike API to consume physical data

3:57 center resources um I actually still this analogy from Raj and he'll talk about it more later um but I love it so

4:03 much I'm I'm frontloading it here um because the abstra design cycle kind of mimics the things that we've been doing

4:09 in network engineering for years right when we're standing up a new data center we design it first and that normally

4:15 takes place oh my gosh we found a room without whiteboards okay this is perfect for what I'm about to say because we've

4:20 moved the design off of the Whiteboard and into the orchestrators and automation tools so now instead of

4:26 drawing stuff up with markers and getting your hands dirty you can actually design and pre-stage the

4:31 network you're going to build um after our designing and planning phase and this as well is where we populate all

4:37 the the attributes and assets that we need to design a data center like asns and and vlans and IP addresses all of

4:43 that goes into our design phase so it's more real than you would ever have in your Vis or your whiteboard designs once

4:50 you've got the the mockup the the design in a state that you're happy with you can go ahead and assign physical

4:56 resources to that design and that's the first time that you ever have to tell abstra about Hardware so you can do all

5:03 of this work in the abstract and then start picking the physical devices that you're going to plug into each one of these roles they can be Juniper qfx or

5:10 they could be any of the other vendors that we support this is the time where abstra decides kind of what config it's going to cook when it instantiates that

5:18 network uh and then finally once we've got our design locked in once we've got all of our physical Hardware uh kind of

5:24 chocked into Its Spots then we can actually deploy the network and we use ztp to to take a a pile of brand new

5:31 switches and bring them up into the configured fabric design that you've you've specified in that day Zero

5:37 section um finally there's the the long tail of these things right because designing and building is kind of the

5:44 the big monumentous event it's it's kind of like a celebration we smash a bottle of champagne over our new Data Center

5:51 and then we go ahead and do the day-to-day work that carries on for years on end and so uh we'll finish up

5:57 today's session talking about how we we embed a whole bunch of probes and monitoring tools into the deployed

6:04 networks um to make sure that they're always running in the state that we wanted them to be in now why is this

6:10 critically important to AI I mean you could convince me that it's no fun to configure uh a clo fabric by hand um but

6:18 in AI when we talk about the scale of deployments that we need uh in terms of both complexity of the Fabrics the

6:25 number of fabrics as well as the demands of the workloads this is not something thing we want to do by hand and we want

6:31 to benefit from all of the research and the lessons that have that have been learned from our Labs from our partners

6:37 Labs um and so this is where the AI jvd takes all of those concepts of abstra

6:42 and bakes those into an AI specific incarnation of these deployments uh we

6:47 do that for example by embedding those rail based designs that prle talked about into our reference architectures

6:53 so you don't have to figure this stuff out by hand anymore it's already baked into the blueprints you load the jvd

6:58 into your abstruct and then you can start from that foundation and tweak and tune it to to match what you want it to

7:05 be um the deployments uh well again uh we're pre-staging all of the uh

7:11 configuration for a typical evpn data center but we're mixing in all the sugar and spice required to make it an AI or a

7:19 data center that can carry AI workloads uh and that means things like that DC qcn configuration and all of the sort of

7:25 the sets of configuration that make um uh a fabric reliable enough to run an AI

7:31 workload that gets baked into the design as deployed and then finally we've built

7:36 a set of probes and optimizations that monitor all of the various cues and flows in a network and can optionally

7:44 tune that network based on its observed values see this front Network backend Network and

7:52 the GPU back and network is separate distinct entities or they all kind of wrapped into one deployment or how's

7:59 this place out so we give it to you as one big Omnibus and then you can break that apart and pick which parts you're

8:05 going to deploy so rajo show you when we spin up the AI jvd you end up with the front end backend and storage designs

8:12 like blueprints with an abstra those are three separate Fabrics that you're deploying they could be on three

8:18 separate independent pools of Hardware um and you can decide and deploy how or

8:23 decide how you want to deploy those independently so it's not like a One-Shot deal you pre-stage all three of

8:28 them or as many as you need and then you can decide how you want to deploy those the terraform that we provide is

8:36 basically like a starting point you can take that and you know like every other project I worked on you take code that's

8:42 already working and hack it to you know suit your needs right so we provide the terraform that gives you like a baseline

8:49 uh that you can start with and then what you do from there is entirely up to you mean it's your

8:56 hardware and the terraform is something you would store into GI Ops or GitHub or

9:01 something like that so you can monitor its configuration changes and things yeah 100% yeah so the demo I'm the demo

9:07 that I'll be showing uses GitHub um uh there are other videos we have online

9:12 where like we used uh githubschool

9:30 we are creating the terraform and that's uh We've created an abster terraform provider to expose all of the the first

9:36 class objects within abstra to terraform and then we consume those resources via a set of projects that we make for each

9:43 of those networks uh and then we publish those uh to GitHub so oh sorry yeah I was going to say like uh the whole um

9:51 intent based approach of Abra lies in very nicely with the declarative approach of terap right so like it's a

9:57 nice match uh because uh you know it's it's not it like like Chris marget you

10:03 know the other guy who wrote the code some of you know him um like he says you know it's verbs sorry it's nouns not

10:09 verbs right so it fits very nicely with the fact that uh in abstra you express

10:15 an intent and you expect abstra to take care of it so we wrote the terraform provider to match match abstra objects

10:22 so this happens this happens and the network does its thing yeah absolutely that answers your

10:29 question well there's a tiny bit more unpacking I want to do on that as well um because and there's a great set of

10:36 screenshots here that kind of proved my point um operating these types of

10:42 networks by hand is not a lot of fun uh I love having these types of designs because it's I hate drawing Network

10:49 diagrams and it's great to visualize kind of an as built or visualize a topology it's really easy to explore a

10:55 design and kind of at a glance figure out what's going on but if you needed to change anything in here or maybe make a

11:03 bulk change to what's going on in your deployed AI data center it would be just

11:08 monstrous to have to click through and to to manually make this change even with a tool like abstra and so we view

11:15 terraform is kind of the power tool that connects to the power tool the universal remote if you will um its job is to let

11:22 us do bulk operations to work at aidc scale with these automation tools and

11:28 again it's with the happy Coincidence of meeting that expectation of our Cloud users they

11:35 want to automate their infrastructure without waiting for us to do our job this is the way that we can expose those

11:41 interfaces in a safe fashion as well um so yeah I mean not just for

11:46 buzzword Bingo purposes but yes this like we benefit greatly from having configuration represented as plain text

11:53 um by having the ability to check that into Version Control and then to perform like testing and validation around this

11:58 as well right ultimately we are running a lab so if we need to throw this at a test cluster and spin something up quickly it gives us that dyn uh

12:06 dynamicism um and yeah sorry I I think we kind of

12:12 talked around this as well but um the the AI jvds this is this is actually a screenshot of what we get when we deploy

12:18 the AI jvd right off the bat in abstra the neat thing about like I said uh whiteboards um you can draw on them

12:25 regardless of what you have right there's no like you don't need to physically have the thing that you're drawing same thing with abstra you can do the

12:31 design phase you can load all of this into an abstra instance and see what the Network's going to look like model it

12:37 tweak it until you click build it doesn't assume that you have Hardware so there's some neat stuff you can do when

12:43 you talk about I think there was a question earlier about um kind of building a digital twin of your infrastructure trialing changes before

12:50 uh you take them into production yeah you can stage it in abster you could also Point your abstra at a whole pile of virtual switches and run it for real

12:56 so there's a whole bunch of really neat stuff you can do once you have these tools and are you able to Overlay this

13:04 say I have my digital twin of the existing Data Center and I want to deploy this as a new section of that but

13:12 utilize the existing software I'm able to try to perform that merger in my

13:17 digital twin it's not a it's not a first class object and abstra you can absolutely do

13:23 that by having a set of switches or a set of Hardware that are virtual junos or virtual junos Evo or and like any

13:30 vendor virtual switch right so you can add that digital twin if I'm understanding your question right you could add a digital

13:36 twin my existing Data Center and I'm add an AI cluster to this and connect it to

13:43 the exist and utilize the existing infrastructure Hardware yeah you can

13:49 absolutely do that it's just it's not there's no first class objects to do that so we don't make any distinction between virtual switches and and

13:55 physical ones you can absolutely plug them together right

14:00 where does security fit into all this stuff uh I mean it's it's kind of a part

14:07 of everything right so as we're as we're exposing apis that puts uh an incredible

14:12 burden on securing the API surface um thankfully we wrap things up in TLS and do search checking um uh I mean like the

14:20 keys to abstra become very very important they're now your API Bearer tokens so um kind of the the if anything

14:29 the the thing that I like about having a tool like terraform and wrapping up your workflows in gitops or something similar

14:36 is you can bake in checks balances and controls so say like sign offs on check-ins right you can have a change

14:42 reviewed to your terraform project before it's pushed live um that gives you mechanisms to control or to to add

14:49 more constraint to a deployment uh it's not necessarily infrastructure security

14:55 but it's it's uh assurance and resilience for your network

15:04 uh we're going to talk a little bit about and I think I talked about this a little right like why we build tools on

15:10 top of abstra right it's to give folks that power tool to do kind of more than

15:15 you would ever want to manually provision um it's also to meet the expectations of those folks coming back

15:21 from the public Cloud right we need to offer that I Can't Believe It's Not AWS type

15:26 experience um there is actually something else we get when we start wrapping Tools around

15:33 abstra we can make those tools talk to each other as well and that's what we're going to show you a little bit about

15:38 today um yeah sure we've created our terraform provider published that up on the internet for you all to try and you

15:44 can all build your own automation around this but that's still kind of an expert level tool so how do we build tools on top of the tools and this is where we'll

15:51 show you and Raj has done a great job building out some integration into service now so we can show you the

15:56 ultimate in kind of consumable interface where you can order up your infrastructure just the same way you

16:02 would order any other it service and I mean it feels strange but when we talk

16:08 about sharing a constrained resource like an AI lab full of gpus there's a lot of people waiting in line to try

16:13 stuff having something like a ticketing system wrapped around this so someone can say you know I get next or put their

16:19 coin on the juke box or take their ticket out of the ticket dispenser whatever your your favorite analogy is this gives them a mechanism to command

16:26 the infrastructure via that highle interface um what we're specifically going to show

16:32 you is how we share better at Juniper um

16:37 specifically how we share our AI lab better um so we have built a set of

16:43 terraform projects that a lot based on our AI jvd that match the the typical

16:48 topologies that we deploy in our lab different over subscription ratios different configurations uh and also we

16:55 have the ability to slice our clusters smaller if needed to share amongst um uh the folks who want to use it uh

17:02 we've gone ahead and built a service now workflow on top of this to create these

17:08 kind of first class objects in service now for the various data centers that we're going to deploy and so when a user

17:14 kicks off a service now request it'll call a set of terraform projects that'll

17:19 reach out to abstra and instantiate the type of network that that user is

17:24 requested and is that plug into the slurm orchestration for the for the actual work that runs across that

17:30 Network or how does that work so we didn't do it for today uh but you absolutely can and that's the funky

17:35 thing about terraform um terraform being multi-resource uh automation toolkit you

17:41 can throw in the automation to stand up the server so we could say uh Pixie boot a handful of servers install kubernetes

17:48 or install slurm uh whatever whatever the application is on top of that and that's kind of what we're envisioning

17:54 like if if an organization or actually if this is what we need to do for the lab um we need need to add on that step

18:00 where we actually bring up the server infrastructure right now we're sharing kind of a pre-built cluster and just

18:05 changing the network subscription ratios and queuing parameters around it but as we take this or as we make this more

18:11 advanced um we'll need to actually slice out the hardware and maybe make a new small cluster for like Raj and I to play

18:18 with over the weekend or whatever we need right sharing is

18:23 caring um and again all of this Stacks up on everything you've heard today right so we're going to be deploying

18:29 and and hopefully what you see is everything gets reer and reer leaping off of the slides and we'll show you how it can jump right onto your abster

18:35 instance running in the cloud run uh your laptop wherever it's running but we'll show you how we deploy all of

18:41 those networks and hopefully you get a sense of how we coordinate like front end backend and storage what parameters we expose to let you tweak those and

18:47 modify them um like I mentioned we have tried

18:53 to publish everything on GitHub so our terraform provider is up on GitHub uh

18:58 the AI cluster designs and all of the automation for the newly published AI gvd are also online and I do not expect

19:05 you to copy those links so here's some QR codes and I promise they are what I said they were there's nothing hiding in these QR codes they're just GitHub links

19:12 it's the easiest way to make an entire room of people nervous um so yeah we've up there got published the AI jbds and

19:19 we have the cluster designs themselves so you can go ahead and see the kind of what we've built why we've built it and

19:25 then the terraform projects to instantiate it like I mentioned you don't need Hardware to do this you can

19:30 throw this at your own abstra instance so if you just wanted to to poke around and see what we've done uh you can go

19:36 ahead and do that and if you just want to see what

19:43 types of automation we're building around abstra you can check out these two repos as well um it's got the actual

19:50 terraform provider itself so all of that development is done in open source and uh it's published up on GitHub uh and

19:57 then we also because I the foundation of every Network automation project always

20:02 starts with copy and paste right I can't be the only one um so we have published a set of examples that you too can copy

20:08 and paste from including everything we're going to show you today so if you want to see kind of how real this stuff is or kind of what we think is important

20:15 to expose out of abstra take a look you can try it for yourself um but yeah I've I've made

20:22 enough bold claims now so now I'm going to pass the mic over to Raj and he can show you how all of this actually works

20:28 um right hey folks uh I'm Raj uh I work on the automation team um we are

20:33 responsible for a bunch of automation mostly in the past two years I've been here mostly focusing on abstra

20:40 one way or the other um so I think I mentioned Chris before uh so Chris

20:45 marget uh Bill Wester and I have been focusing on the terraform provider for abstra um the I mean the whole vision of

20:53 this is that you know you should be able to manage your data center just like you manage everything else um the general

20:59 idea is like you want to deploy an application uh you create the terraform like application guys have terraform for

21:06 their application um the idea is like if you could have if if you could have your

21:11 network changes also as part of the same terraform configs uh you can deploy the whole thing in one shot and everyone's

21:17 happy and nobody gets woken up at night um so that's the goal here

21:23 uh so uh like everybody else mentioned like like I said like the whole thing

21:29 Builds on previous things um like everybody else here mentioned right like

21:34 abstra makes it possible with its API and the provider that builds off of the API um to do to basically automate your

21:42 data center uh so what we see here is service no on the left uh terraform

21:47 cloud in the middle and an empty appst on the right right um by the way that is

21:52 a cloudlabs instance uh if you want to try out abstra I highly highly recommend cloudlabs as a developer I I mean I

22:01 don't know what I would do without cloud Labs right you can bring up your instance there are a few uh you know

22:06 there are a few kinds of instances around uh and it's great for like you know just developing code and it's

22:12 abstract Cloud labs for those of you furiously Googling right now yeah yeah it's cloudlabs appstore.com you you'll

22:17 get there um I think it's running at two times the

22:24 speed which is I think appropriate I don't really type that fast um so we

22:30 wrote some custom automation uh in service now that talks to terraform cloud and terraform cloud is effectly

22:36 effectively are you know sits in the middle and it automates that appst out there uh it's going to build it's going

22:43 to basically build out uh the three uh Fabrics that uh that that are in the

22:49 ajvd um and you know it's just going to go right

22:56 um yeah I mean in real time this takes 5 minutes okay and I'm running it at like

23:02 double speed so it's going to take 2 minutes um everything that's here is I

23:07 mean except for the uh the the service now stuff which have not yet put uh put on GitHub everything else here is on

23:14 GitHub um so you know while that's running it's running it takes another

23:21 it's going to take another minute somewhere you mentioned abstra U AI extensions for App Store what sort of AI

23:26 extensions are there for app uh we are in the process of developing those uh basically like uh what we're

23:34 going to show later today uh you know we're going to be doing some um uh we

23:39 have some close loop automation that we're working on so that's that's part of it um uh Jay is going to show you

23:46 some custom collectors we' have created to pull custom Telemetry that are pertinent to AI uh in abstra uh and then

23:53 the demo later today will will basically put the whole thing together to do some tuning for abster uh sorry to tune the

24:00 fabric using abstra for a applications okay not for Abra um yeah so there are

24:07 310 objects this is going to go and it's going to create it's going to create your your air cluster uh sorry your um

24:15 it's going to deploy the a Fabrics into abstra for you um just like in the

24:21 cooking shows I have the completed you know the completed dish out here because nobody wants to watch like you know

24:28 something boiling for an hour so you know and we are in the Julia Child's meeting room so it's just perfect I

24:35 couldn't resist that's perfect I I didn't think of it I'm slipping so you

24:41 know so you go in here like uh you go in here um you have your you know it

24:48 basically shows you like like the the fabric this is basically what our lab looks like except you know this is a

24:54 virtual instance this this does not have like devices backing it but this is what our lab looks like right and I want to

25:00 call back to what uh Nick said before um this has taken you all the way to the

25:06 point where you order your gear you rack it up and you know you actually have to do it right so this would have started

25:14 life in a normal environment on somebody's Blackboard and then onto a vis onto a vis document everybody

25:21 approves it everyone's happy and then like you buy stuff you rack it up you do all of that this has this has basically

25:27 done all that for you right the only thing remaining is to buy really expensive gear and you know make it

25:35 happen get as many gpus as you can plug them in yeah yeah instantly like um the the the

25:44 terraform for the a for the a jvd if you give it a list of device name if you give it a list of device IDs will also

25:51 take care of deploying the devices in like it'll take care of deploying the devices um so you know this is all this

25:59 happened you know basically because of what abstra enables right uh is it possible to do all of this without

26:05 abstra yes but it's going to be really really complicated right the kind of the marriage of intent based um networking

26:14 and the kind of the declarative nature of terraform really lets us do this in very very short you know in very very

26:20 short order right um the idea is to take this Automation and hand it to the end

26:26 user right because ultimately the person who kicked off that service now request

26:31 will not be a networking person and should not be either right like it's going to be the the data scientist

26:38 somebody who wants an environment to play with they do it nobody calls us and

26:46 it just happens right and if you're hearing a note of selfishness in all of this it's because pul just invited the world to come to our AI Data Center and

26:52 we need to be able to spin around a different design and a different deployment for every customer who walks through the door and

26:58 I don't want to be the one doing it I'd much rather write the code put it up on GitHub and let people have fun yeah

27:04 right uh and like the codes right here uh let me let me show you like something

27:10 I did just a few minutes ago I can figure out how to uh so here is like

27:17 here is the there you

27:22 go your tab is hidden yeah um so

27:30 this is a change that I pushed out just an hour ago while sitting here basically all I did was uh you know change the

27:38 name of this blueprint to say like backend

27:43 GPU fabric for cfd right I did that simply by pulling down the terraform changing

27:50 the name here and then pushing it to get and we wanted to show you a really good example as well how we've put parameters

27:57 in the AI jvd to help you control it so I mean it's not just things like the names it'll be over subscription ratios

28:03 it'll be uh Hardware platforms it's all of the details that you would need to change as you're deploying uh your data

28:10 center right like you know you want to partition your environment uh you want to create like a new virtual network uh

28:16 with just for like two or three devices and hand that out to somebody there is terraform for that right and you could

28:22 tie this with an actual application deployment like slurm or something and the whole thing would be taken care of

28:28 like in one terraform apply that's the vision here um I think think I've said

28:34 everything I wanted to say uh any question do you see service now as a as a as a an alternative to terraform no no

28:41 as a as a front end so that folks like once we we're always thinking about what tool are we going yeah so what do what

28:48 tool do we want to hand up to the layer above us so from the network we wanted an API we got abstra from appstor we

28:53 wanted a a text Bas automation tool so we have terraform now as and I can write

28:59 terraform as a recovering network engineer so it's a great highle interface but for the folks who need to

29:04 go one step higher we need a button and service now is the button that everyone expects to provision their company's

29:10 infrastructure so we put the the make the AI DC Go Button in service now uh

29:16 the cool thing about terraform and terraform cloud is you can put that button anywhere so we have customers building this into cicd pipelines we

29:22 have folks using service now we have folks building their own custom dashboards yeah if you look at like uh

29:29 we have other examples like there's another example where we integrated GitHub and terraform Cloud so that we it

29:36 makes it look like like basically just like how a software engineer you know creates a pull request and it gets

29:43 deployed all the way to production did the same thing with terraform and GitHub

29:48 and uh uh abstra right so you basically express your network change as uh as a

29:54 pull request right um and I'm not even a recovering network engineer I was a

30:00 programmer right I've always I've always do like uh uh Network adjacent like the

30:07 the last job I did like we were automating like a private Cloud so I wrote the code I would work regularly

30:12 with the networking guys so it makes it you know even I can understand this right so that that

30:19 that's kind of the power of this whole this whole infrastr this whole ecosystem right and the idea is like the next

30:26 level is like uh is is an end user who really doesn't care about networking or

30:32 about devices or about any of these things all they want is like to log into

30:37 their to their slum right make my application yeah exactly so just go into service now ask for what you want and

30:45 you know your company gives you your SL one service now enables all of their Integrations as well a slack poot that

30:52 says hey I'm a researcher and I want this and it goes up to service now it goes over to and everything kind of gets

30:59 deployed the way that they want it and they have no idea what's on the back end yeah and like by Magic the leg bone

31:05 connects to the hip bone and everything dances right and that's that's what we're aiming for so we actually have some practical examples that Jay's going

31:11 to walk us through we can show you a live Network where we built and deployed all this and how we monitor it and kind

31:16 of what some of this custom AI work that we've done is um so maybe the best way

31:22 to to show how this all works is to to show you just one other high level thing first you this is obviously going to

31:29 work really smoothly in a in a deployed data center where I'm laying things over top of the existing um CLA architecture

31:38 but you're also running a lab here so there's going to be times when you're deploying a lab that's going to require

31:45 recabling of of things because of the the new application that's that's being

31:50 deployed so what what happens on the physical end to to assist the physical

31:55 end when the architecture want to deploy has to be cabled differently than what's out there now will it detect that how

32:03 does it help me get the the cables moved yeah so on the network side it's using ldp to map out where things are and App

32:10 Store will detect any deviation from its set of like it set topology as what it calls an anomaly so it'll say that I

32:16 have a a link anomaly I'm not getting the lp or I'm get not seeing the lp neighbor that I should be seeing on that

32:23 link um the other um the other thing that

32:28 Appstore will do is print out cable Maps uh of a topology so you can say where should all this stuff plug in and it's

32:35 meant to be handed to a data center technician or someone who's going to go plug it in usually me um so yeah that

32:43 highlight the changes then so I've deployed architect I have a deployed I need to deploy B that's going to require

32:51 six cable swaps that my fieldtech has to ride out to the dark Data Center and I haven't seen a diff that's a I may have

32:57 to look into that for you request yeah I just I heard that but I I really do like

33:05 that so are we going to see a demo today of uh say a failed network configuration

33:13 and are we going to see any other types are we going to see specific examples of a model uh an AI model being run either

33:21 in learning mode or in fine-tuning mode Jay's going to show you a whole range of

33:27 anomalies uh it's we've got like a very Network Focus here so we'll show you what congestion looks like or what

33:32 packet loss looks like but Jay I'll walk you through some some live scenarios and then we can torment Jay once he's up

33:38 here and make him show us whatever we want