Jon Mischel, Director, Juniper Networks

Data Center Fabric Management and Telco Cloud | Demo

Telco CloudData Center
Jon Mischel Headshot

Delivering intelligent operations in the automated data center.

Experience the automated data center and telco cloud with simplified Day 0, 1 and 2 operations and cloud-native agility. Unlock the benefits of Juniper Apstra and Cloud-Native Contrail Networking (CN2).

To learn more about our Telco Cloud solutions:

Show more

You’ll learn

  • How to rollout a new network service across thousands of locations in record time

  • Why an intent-based model that is reliable, repeatable and multi-vendor is critical for scaled deployments

Who is this for?

Network Professionals Business Leaders


Jon Mischel Headshot
Jon Mischel
Director, Juniper Networks


0:00 [Music]

0:01 hi I'm John Mitchell and I'm director of

0:03 marketing for Juniper's cloud ready Data

0:05 Center and welcome to mwc Barcelona

0:08 2023. in this demo we're going to be

0:10 focusing on Juniper's automated data

0:12 center imagine that you're a service

0:14 provider we connect and you've got

0:16 thousands of locations around Europe

0:18 particularly here in Spain and you're

0:20 trying to roll out a new video service

0:21 before mwc Barcelona what do you do well

0:26 it's pretty difficult to do if you are

0:28 doing it manually almost impossible to

0:30 do anything like that at Telco scale so

0:32 we're going to use our automated data

0:34 center Tools in order to simplify the

0:36 deployment and make it easy for you to

0:38 get these these products out the door

0:39 the automated data center is built

0:42 around a intent-based model so that you

0:45 don't have to worry about CLI it's

0:46 reliable so it's repeatable and it's

0:49 multi-vendor so you can decide on the

0:51 products that you want to deploy based

0:54 on let's say Port density performance

0:56 cost or maybe it's a supply chain issue

0:58 and you need to look at an alternative

1:00 vendor so what I'm going to do in this

1:01 demo is I'm going to walk you through

1:03 developing a rack deploying virtual

1:06 networks across a multi-vendor rack and

1:09 then how we deploy Telco Services across

1:11 a high performance server and do it in

1:14 record time we're going to start off

1:16 with deploying app store app stores are

1:18 multi-vendor data center IP fabric tool

1:21 it allows you to take what would

1:24 normally take days possibly weeks and do

1:29 it in minutes so we're going to deploy a

1:31 rack

1:32 so here we have we connect trying to

1:35 deploy a new video service and they need

1:37 to add racks to their different

1:38 locations so again we connect as data

1:41 centers they have Telco clouds they have

1:43 Edge clouds all throughout Spain if

1:45 you're going to do something like this

1:47 it's possible to do it manually but it's

1:50 impossible to do anything at Telco scale

1:52 when you're dealing with CLI I'm going

1:54 to show you how Appstore does it we

1:56 currently have in this particular

1:57 location a two-spine environment with

2:01 one leaf what we want to do is we want

2:03 to add two new leaves to this and we're

2:05 going to use templates in order to do it

2:07 so you don't have to configure this

2:09 manually you can define those those

2:11 templates those templates are

2:13 incorporated into and delivered through

2:15 App Store and we can apply them without

2:17 a whole lot of product knowledge so we

2:19 go to racks we're going to go add rack

2:21 and here we're going to select from a

2:24 set of predefined templates the type of

2:27 rack that we want to deploy in this case

2:29 it's an evpn ESI rack which is kind of

2:32 like multi lag so here we're adding the

2:35 two leaves we have a dual connected

2:38 server and then two single connected

2:40 servers now we're going to add this into

2:42 our existing environment and Stage IT

2:45 next we have our topology so now we've

2:49 added the two leaves to our existing

2:51 environment and we need to drill in and

2:54 start to add the products so at this

2:56 point we just have the topology we

2:59 haven't picked any platforms yet so what

3:02 we're going to do here is start with the

3:03 interface map the interface map defines

3:05 the number of ports the type of ports

3:07 are these uplinks or the axis interfaces

3:09 it's not defined based on vendor type so

3:12 in this case we're going to go in we're

3:13 going to select from a set of menus and

3:15 you'll have these many drop down menus

3:17 that allow you to pick the type of

3:19 interface map that you want in this case

3:21 we're picking a qfx and by default it's

3:24 going to pick the platform as a qfx

3:26 itself next everything gets staged so in

3:29 this case we can batch everything so

3:31 theoretically you can build your

3:33 topology have it ready to go but you can

3:36 do it with without having a single

3:37 product in-house this allows you to

3:39 streamline the process of getting your

3:41 deployments ready and once that

3:43 equipment ships on site you can deploy

3:45 so here Appster goes through a check

3:48 process where we validate the

3:49 configurations we commit those

3:51 configurations to a live Network and we

3:54 converge so you're ready to go now we

3:57 have the topology but now we're going to

3:59 add virtual networks to this environment

4:01 these virtual networks are going to add

4:03 compute storage and networking so that

4:05 we can extend our service to a broad

4:09 customer base we're going to do this

4:10 without any CLI so when you're adding

4:13 any type of of network or racks to your

4:16 system there's a whole lot of things

4:18 that you have to do you've got to Define

4:20 your topology and your connectivity

4:22 you've got to Define Network IDs and IP

4:25 addresses get the design your routing

4:27 environment and then you have to type it

4:29 out but not if you use an app store apps

4:31 it allows you to do this without

4:32 understanding or knowing clis which is

4:35 which is great when you have

4:37 environments where you're not sure which

4:39 products are going to be in demand and

4:40 you're not sure what expertise that you

4:42 have in-house in order to deploy the

4:44 product

4:45 let's add this capacity and we're going

4:46 to use apture to do it so we're back in

4:49 another environment where again we have

4:51 two spines we have three leaves

4:53 this is a multi-vendor rack so we're

4:55 just gonna drill down a little bit into

4:57 this existing deployment and we see that

4:59 we have an Arista pair on one side

5:01 [Music]

5:02 and then we're going to drill down into

5:03 Leaf three and we can see that this is a

5:06 Cisco device

5:08 so now we're going to Define our virtual

5:09 Network

5:11 so in this case

5:13 we go to our ad virtual Network screen

5:15 so you can see that we've got a name for

5:18 the virtual Network it's Finance we're

5:19 assigning it to an existing routing Zone

5:21 kind of finance and I'm going to leave

5:23 these two entries blank right now the V

5:26 and I and the VLAN ID just to show you

5:28 how Appstore goes through to validate

5:29 configurations before they're deployed

5:31 we're going to pick the different types

5:33 of configurations for this particular

5:35 deployment the ATP Services V4

5:37 connectivity Gateway addresses Etc is it

5:40 tagged or untagged and then we're going

5:42 to Define which switches we actually

5:45 want this virtual Network to be applied

5:47 to if you're building a virtual Network

5:48 you don't necessarily want to send it

5:50 out Rogue to every single switch you

5:52 want to be specific so you can kind of

5:54 improve your security and reduce your

5:56 your blast radius so here we're going to

5:59 put it on both switches but we can get a

6:01 little more granular in the next set

6:03 where we can then Define the ports we've

6:05 added our virtual Network so we can see

6:08 its Finance we have our vxlan but you

6:10 see something shows up as red red just

6:13 means that the there's a configuration

6:14 that's incomplete or there's an error so

6:17 here we can see we have to add the vnis

6:19 so we're going to pick from a pool we're

6:22 going to have the v-tep again we're

6:23 picking from a pool and then we can go

6:26 in and Define which ports of those

6:28 switches that we've defined are going to

6:30 actually have that virtual Network now

6:32 we're staged we want to take a look at

6:34 the configuration so we're going to

6:36 drill in a little bit on

6:38 in this case an Arista switch now I

6:40 don't know anything about EOS after has

6:42 rendered the entire configuration based

6:44 on the intent that we've defined leading

6:46 up to the stage so there's no CLI work

6:49 apps are renders that configuration

6:51 it'll do the same thing for iOS it'll do

6:53 that obviously junos and others we then

6:55 stage it and we commit it to the network

6:58 once it's accepted the network is

7:00 converged we turn green

7:02 we want to add a Telco workload now Toca

7:05 workloads a lot of times require very

7:07 high performance so these workloads need

7:09 to be deployed in an sriov host so these

7:12 workloads plug directly into the IP

7:15 fabric in order to connect those

7:18 workloads and those networks in the

7:20 underlay to workloads that are deployed

7:23 by kubernetes and through the the

7:26 overlay networking and sdn for Juniper

7:29 that's our Cloud native control

7:31 networking or cn2 which is again it's an

7:34 sdn and cni built into one

7:37 you need to manually Stitch those

7:39 together up until now so what we've done

7:42 is created a cn2 and Abstract plug-in

7:45 and this plugin allows you to stitch

7:48 these ships in the night networks

7:49 together without creating troubled

7:51 tickets or doing things manually which

7:54 can result in errors so let's walk

7:57 through adding in this case a charging

7:59 function for this video service that

8:01 we're deploying here in Barcelona so

8:04 first we're going to take a look in app

8:06 store we're just going to show that

8:08 there's an existing rack but we don't

8:10 have any existing virtual networks and

8:13 we don't have any existing routing zones

8:16 so it's a blank slate

8:18 so now let's go into the web GUI the web

8:20 GUI lets you define the interface into

8:22 appstra so you put in your credentials

8:25 the location IP addresses the blueprint

8:28 that you're going to be using and this

8:30 instantiates the plugin so that abstra

8:33 and cn2 can communicate with one another

8:35 to actually automate the the building

8:37 and stitching of those virtual networks

8:39 that are both SRI OV hosted and not SRI

8:42 will be hosted so the first thing we're

8:44 going to do in again the cn2 GUI is

8:46 we're going to find the network

8:48 attachment definition now this is the

8:50 same network attachment definition that

8:52 you'll see with multis and cni and we're

8:55 looking at net 20. we're just going to

8:57 drill down in here we're using it we

9:00 have a yaml file and we have our own

9:01 embedded animal editor and we can Define

9:04 what that Network would look like so

9:06 that when workloads are brought up on

9:08 SRI OV we can either dynamically build

9:10 the network to attach that SRI OV host

9:13 or we can attach that new pod to an

9:17 existing Network

9:19 now here we're going to look at pod 20-1

9:21 and we're going to open it up at this

9:24 definition we're going to augment and

9:25 we're going to add in it the NAD

9:27 definition this definition becomes a

9:29 trigger so that when the Pod is created

9:32 abstra sees that net 20 references the

9:35 plug-in itself so that the network is

9:37 automatically built so now we can drill

9:39 down from cn2 to look specifically at

9:43 the state event Network we've created

9:45 the Pod the network is existing here in

9:48 cn2 but the network is also existing and

9:51 connected in app strip so we can see all

9:54 the state all the connectivity and

9:56 everything is working and everything was

9:58 automated we didn't have to issue any

10:00 trouble tickets or do any manual

10:02 configurations in order to stitch these

10:04 two networks to wrap things up Juniper's

10:06 automated data center it's intent based

10:08 so again you don't have to worry about

10:09 clis in order to operationalize a

10:11 network it's reliable meaning that it's

10:14 repeatable so the processes are there

10:16 they're baked into the the workflows and

10:19 it creates consistency in your

10:20 deployment models and it's multi-vendor

10:23 so that you can pick products based on

10:25 your criteria

Show more