Reid Stidolph, Technologist and Senior Product Manager, Juniper Networks

Experience the Juniper AI-Driven SD-WAN Demo

Demo Drop AI & MLSD-WAN
Reid Stidolph Headshot
Slide content has three headings across the top that say, “Optimized for assured experiences. Proactive AI-driven operations and support. Open and programmable cloud.” Below that is a network diagram. At the top is “Juniper driven by Mist AI” and “Marvis Virtual Assistant. Text below that says, from left to right: “DEVICES AND USERS, Wireless, Wired, WAN, CONNECTED SECURITY, MULTI-CLOUD APPS.”

Experiencing is believing: It’s so easy to onboard a Session Smart Router

As part of Networking Field Day 27, Reid Stidolph and Adam Morris get their hands dirty with a hands-on demonstration of key elements comprising Juniper’s AI-driven SD-WAN. 

Show more

You’ll learn

  • How to onboard and configure a Session Smart Router device

  • How day two operations are simplified with Service Level Experience metrics for the WAN that describe user experiences across your SD-WAN deployment

  • How to ask Marvis to identify elusive issues and automate troubleshooting

Who is this for?

Network Professionals Business Leaders

Host

Reid Stidolph Headshot
Reid Stidolph
Technologist and Senior Product Manager, Juniper Networks
Adam Morris Headshot
Adam Morris
Systems Engineering Manager, Juniper Networks 

Transcript

00:08 you know as as we go into the demo

00:11 uh

00:13 session smart routing we as we mentioned

00:15 it's been out there on the market for

00:16 seven years now being proven and and

00:18 being improved uh with that tunnel free

00:21 approach

00:22 um but as sudhir alluded to there's

00:25 really another side to this and and that

00:27 and that's missed um you know missed as

00:30 as this technology and everything was

00:32 coming to market and and being deployed

00:35 uh with 128 technology you know mist was

00:38 actually on the exact same mission as as

00:40 the ssr to ensure a better user

00:43 experience so

00:44 you know while the ssr has been kind of

00:46 coming at it from the wire up mist has

00:49 really been at it for a while now um

00:51 from the cloud down

00:53 and

00:54 nowhere are the operational challenges

00:56 resulting from

00:58 you know from

00:59 these changes in the network uh felt

01:02 more acutely than the far edges the

01:04 remote branches the the offices the

01:07 enterprise campuses and you know

01:10 as sudhir alluded to at juniper with

01:13 mist um you know we've long been

01:15 applying the innovative modern cloud ai

01:18 solutions to these challenges with wired

01:20 and wireless access and so now with wan

01:22 assurance

01:24 powered by session smart routing really

01:25 the voices the forces have joined and

01:28 and we're super super excited about that

01:31 and uh want to show off a little bit of

01:33 what that looks like um so

01:35 with that i've got some uh some portions

01:38 of demo here that are are

01:40 recorded pre-recorded and then i've got

01:42 the demo environment uh live at my

01:45 fingertips so we can bounce around a bit

01:47 uh through this but

01:49 um

01:50 you know this this is what the branch of

01:54 the future looks like um here we have

01:57 our

01:58 uh juniper switch

02:01 and our

02:02 ssr 130 device and a

02:05 a and an

02:07 um so

02:09 you know kind of wanted to walk through

02:11 the

02:13 you know the lifespan the the journey of

02:16 deploying and operating this so let's

02:18 start in the mist dashboard where we

02:20 have our sites configured for an

02:22 enterprise organization called acme

02:24 we have sites here at westford and

02:26 sunnyvale which are the hub sites and

02:28 cupertino is acme's brand new site that

02:31 we've gotta bring online

02:34 in the organization-wide settings we now

02:37 have the lan uh we see the wan here

02:40 starting and starting with networks you

02:42 know this is where we describe who our

02:44 user populations are and the segments

02:46 that we're going to use to connect them

02:48 in the land so i've got a segment called

02:50 trusted here that's going to describe

02:52 you know just broadly our corporate

02:54 users

02:56 so then we move to services you know

02:58 this is where we describe the networks

03:00 and the applications that our users in

03:01 the land are going to connect to

03:03 services describe the destinations

03:06 of network sessions

03:08 uh for

03:09 you know and they and they each have uh

03:11 discrete routing access

03:13 uh and security policies so here we see

03:16 you know one that's very broadly defined

03:18 this uh internet service

03:21 um just defined by a quad zero and then

03:23 we see some that are defined by built-in

03:26 apps uh that are uh detected natively uh

03:31 within uh within the sd-wan

03:34 and uh again there's a whole

03:36 uh

03:38 a variety of of built-in apps um those

03:41 apps are also categorized so you could

03:43 create a service ultimately uh based on

03:46 an app category or customize uh should

03:49 you need to go that route so

03:52 you know we have a few services that the

03:53 wan is going to deliver

03:55 we have a network tenant describing

03:57 who's going to be accessing

04:00 you know now we need to describe a

04:01 little bit about the network the

04:03 topology that's going to connect them

04:06 so for our

04:08 acme or for our cupertino site we have

04:10 mpls and broadband for connectivity

04:13 and we've defined each of those as a set

04:15 of overlays

04:18 that are going to connect to the

04:19 westford and sunnyvale hub sites

04:23 so as we move into our wan edge

04:25 templates and see some of the local

04:27 configuration that gets applied to the

04:29 site

04:30 uh

04:31 we we see here that uh

04:34 um you know these these templates are

04:36 ultimately how we can deploy one site or

04:40 thousands of site ultimately using these

04:41 templates and and you can see here that

04:43 our uh sunnyvale site is is hub

04:47 um and our cupertino site is spoke so

04:51 just some simple topology configuration

04:53 there uh and then on our lan uh we kind

04:57 of like we described earlier where we're

04:58 just saying that this lan interface uh

05:01 is connected to the the wired and

05:03 wireless access network um

05:06 and uh

05:08 and it's our trusted network segment and

05:10 then here you see a little bit of path

05:12 preference um

05:14 being configured where we're going to

05:16 kind of

05:17 set some default strategies for using

05:20 the available mpls and broadband

05:22 connectivity

05:23 so at this point we have our who who our

05:25 user population are is

05:27 uh we have our what what the service the

05:30 services that the wan is going to

05:31 deliver and a strategy for how that

05:33 connectivity is going to be used and

05:35 this service policy is really where we

05:36 tie it all together we declare here that

05:39 our trusted users are allowed access to

05:41 the internet service

05:43 and it's going to be delivered with the

05:45 path selection policy we defined that is

05:48 mpls first

05:50 and broadband second so that's it that's

05:52 it our basic you know basic wan design

05:55 uh for our cupertino branch is complete

05:58 uh

05:59 and you know we have our wired and

06:01 wireless access

06:03 devices similarly defined and really

06:06 the last thing we got to do is connect

06:08 them so you know that

06:10 fast forward in your mind here um just a

06:13 second say so this is day one um um this

06:16 is the big day where we turn up our

06:18 devices turn up our cupertino branch

06:21 they've arrived on site and we're going

06:23 to take this branch live so you know

06:25 here we're seeing

06:26 our ssr 130 device it has a missed claim

06:30 code on it

06:31 this is

06:32 consistent with what you would have in

06:34 the missed access point and the juniper

06:36 ex switch

06:38 you know fresh out of the box

06:40 so now we go into our wired and wireless

06:43 device inventory we've already claimed

06:45 the switch in the ap

06:47 many of you may be familiar with that

06:48 process of claiming it's real simple um

06:52 and and we're going to do something

06:54 similar here with the wan edge now so

06:56 enter a simple claim code you know it's

06:59 worth noting we're going to sign it to

07:00 our cupertino site

07:03 and it's worth noting that as we claim

07:05 this device we're showing it here sort

07:07 of one at a time

07:09 but in reality these can be done in bulk

07:11 most often they're done in bulk as the

07:12 pace of the deployment wraps up so you

07:14 can do these you know hundreds or dozens

07:16 or hundreds at a time if needed

07:19 so the device is claimed it instantly

07:21 shows up in inventory and uh now we just

07:24 gotta have the person at the site uh

07:26 plug these things in and power them up

07:29 for the first time so

07:31 uh

07:32 as these are plugged in here we got our

07:34 broadband and our mpls connectivity blue

07:36 for broadband yellow for mpls

07:41 and

07:42 the last step is really just uh applying

07:45 the power so power connectivity

07:49 and uh

07:51 as this uh as this happens power is

07:53 applied and the devices boot up and they

07:56 all

07:57 reach out and try to connect to the

07:58 cloud so one by one they learn of their

08:00 function

08:01 uh and and their topology and

08:03 configuration and transcend transition

08:06 into service so the wan edge provides

08:08 connectivity for the switch provides

08:10 connectivity for the ap and last but not

08:12 least we see our device here

08:15 uh you know the the ssid for a corporate

08:18 network just starts broadcasting it

08:20 connects to it and we're online um so as

08:24 we uh you know prove that we're online

08:26 here with a few uh

08:28 hits to a web page

08:30 um we bounce back quickly to our um

08:34 missed dashboard

08:37 and we're going to see now in our

08:38 inventory that all our devices that

08:40 we're in inventory are all connected

08:42 they're all operational uh

08:44 data is starting to flow in from uh the

08:47 uh the telemetry from all three of these

08:50 devices and we're online so you know

08:52 more than just when this really is a

08:56 complete branch ztp experience so wired

08:59 wireless lan um it's it it is how

09:02 thousands of sites that like sudhir

09:04 mentioned can be rolled out in a

09:05 frictionless

09:08 manner and uh again

09:10 fast forward your mind a little bit this

09:12 is this is you know the the site's been

09:14 live for a few days um and uh we want to

09:18 kind of look at some of the wan insights

09:20 that are being observed by mist

09:23 so

09:24 at our insights page we get a view into

09:27 the land performance of the site we we

09:29 see a historical view of the land

09:31 related events and and we see various

09:34 applications that are being consumed

09:35 over the land

09:37 uh and also visible is a list of active

09:40 clients that are utilizing the land for

09:42 connectivity to services

09:45 and you know really a comprehensive view

09:47 of the user activity at the location

09:50 and uh

09:52 you know further down we're gonna see

09:53 some

09:54 uh wanted device performance data

09:57 plotted across the set of charts we see

10:00 you know cpu memory and and then we see

10:03 some of the the uh

10:05 uh the path performance um data that's

10:08 being monitored measured um by you know

10:12 in the various paths uh that

10:14 that this device uh can see so um

10:19 you know

10:20 it's great to have a detailed page but a

10:22 detailed view into an individual site

10:24 but one really nice thing about the the

10:27 uh the miss dashboard it kind of takes

10:29 it back a level gives you this top level

10:31 view of all the sites and it starts to

10:34 boil all that telemetry data down into

10:38 something we call sles now those that

10:41 have used mist for wired and wireless

10:44 may be familiar with sles for those that

10:47 don't or haven't sles stand for service

10:49 level experience

10:51 so looking at our at our cupertino site

10:55 we can see that our service level

10:57 experience this is a measure of user

10:59 experience

11:00 uh and it's a little less than than

11:03 perfect so you know clicking into the

11:05 site we get a a closer look at these

11:08 sles they're broken down into three

11:10 important health categories

11:12 uh that play a role in user experience

11:14 the wan edge device health uh the health

11:16 of the wan links and paths

11:19 and the health of the applications

11:21 themselves so

11:23 each sle is is really a simple measure

11:26 of user experience called the user

11:29 minute and you know simply put what this

11:31 is telling us is what our user

11:33 experiences on the lan are per user

11:36 per minute and

11:38 you know behind these simple

11:40 measurements are the complex and

11:42 powerful models of of the miss cloud fed

11:45 by the rich telemetry uh of this session

11:48 smart network um so you can see for for

11:51 each sle we get a breakdown of root

11:53 causes of the issues that have been

11:55 identified

11:56 and you know whenever a user experience

11:59 is poor on the land

12:01 uh miss tells us about it and we it

12:03 tells us not only tells us about it but

12:05 tells us the root cause

12:07 so here we're going to look at a you

12:10 probably saw the screen flash quickly

12:11 there we're going to look at a uh

12:14 another example of these sles in action

12:16 you know here we see where our gateway

12:19 health is healthy our wan link is

12:21 looking great

12:23 but our application health is a little

12:25 low

12:26 so

12:28 we dive in a little closer

12:30 and this is a real example by the way

12:32 recorded a few months back some of you

12:34 might remember this

12:35 but we look into our application

12:37 services and we see

12:39 mist is telling us that some

12:40 applications are slow slow to respond

12:44 so let's click into it

12:46 and uh and let's get a sense for what

12:49 actual what are the affected items by

12:51 this

12:53 here we see a list of applications a

12:55 note ring on that list is a is an

12:58 application that we're noticing missed

13:00 is noticing some slowness in

13:04 so so this is

13:07 let me pause for just a second here so

13:09 this is missed telling us that the wan

13:11 is fine

13:13 the wired and wireless is fine

13:15 the device itself is fine but

13:18 applications themselves are being slow

13:21 seems a little

13:23 too good to be true so let's let's go uh

13:25 let's go see if there's any report of

13:27 this uh

13:28 um

13:30 out in the public news and sure enough

13:32 on the day that we recorded this video

13:34 or this demo um

13:37 there was a amazon server outage that

13:39 was causing um a number of sas

13:42 application issues and the ones that

13:45 were shown you know that were in the

13:47 headlines were the exact same ones that

13:49 were that were seen in uh in the mist

13:52 dashboard so a really great example of

13:54 how

13:55 um you know how these sles basically

13:58 translate into

13:59 uh into root causes of issues that are

14:03 impacting user experience even when

14:05 they're outside the land um and and you

14:08 know that that really saves on

14:10 troubleshooting time and getting to root

14:12 causes

14:14 so that's a little bit about sles uh now

14:17 let's get into marvis um

14:19 again marvis is is familiar to many that

14:23 have that have uh uh it's been around a

14:25 while in the wired and wireless space

14:27 and of course now

14:29 marvis is seeing land data which is

14:32 great and so here we have marvis actions

14:35 uh marvis actions are the proactive side

14:38 of marvis this is where it identifies

14:41 actions that can be taken to improve

14:43 user experience so if there's an action

14:45 that a user

14:46 an administrator can take to improve the

14:48 network it'll be brought to the

14:50 forefront here

14:52 for our when in this example we see that

14:54 marvis has identified a persisting lte

14:57 quality issue

14:58 and here we're drilling into some of the

15:00 details of that issue to get to get her

15:02 a better sense of of what can be done

15:04 about it so um you know a really great

15:06 example of marvis helpfully suggesting

15:09 that

15:11 maybe we should go to this site and uh

15:13 and adjust the the antenna a little bit

15:17 so that's marvis actions

15:19 now

15:21 marvis isn't just in the background

15:24 working on sles um

15:26 and uh

15:28 looking to suggest helpful actions

15:30 marvis is ever present in the foreground

15:32 of the missed experience as well

15:35 you know ready to answer questions about

15:37 the network at any time

15:39 you can ask it to do things like

15:40 troubleshoot a device

15:43 access documentation

15:47 and a variety of things about the wired

15:49 and wireless network and again truly now

15:52 about the when as well so at our cooper

15:55 site cupertino site we have we know

15:57 teams is an application uh that's an

16:00 important collaboration app

16:02 and one of our users arsteidolf me

16:06 has been uh reporting some bad

16:08 experiences with team calls so let's go

16:11 ahead and ask marvis about that

16:14 and just like that marvis quickly

16:16 responds with a handful of team sessions

16:18 that it determined were teams calls for

16:21 our users yesterday so let's great let's

16:23 let's ask marvis to go further let's

16:25 troubleshoot it

16:26 and here we see an end-to-end path that

16:29 the session traversed

16:31 wired wireless and now when and looking

16:35 further into the wan we see

16:37 it has

16:38 shown us

16:39 that there was high network jitter

16:42 experienced in the land that most likely

16:45 impacted

16:47 this call

16:49 so think about that a moment

16:51 a simple question you know why was my

16:53 team's called bad and a question that

16:55 would historically need to be answered

16:57 by top technical operators across you

17:00 know each device uh you know for wired

17:03 wireless

17:04 and when going device to device pouring

17:06 through logs and packet captures

17:09 mountains of monitoring information just

17:11 to answer where the session went

17:13 and where it went wrong and you know

17:16 here we have a simple question being

17:18 answered simply by marvis

17:21 so now that marvis pinpointed the issue

17:23 let's actually go to our sle's at the

17:25 land site and and uh and sure enough

17:28 when we go to yesterday

17:30 we we see the uh the root cause uh that

17:34 was identified by marvis is corroborated

17:37 by our sles here

17:39 in our affected items we can see the

17:42 clients

17:43 um that were affected and we see that

17:45 this jitter instance of jitter affected

17:47 all of our peer paths so it uh it was

17:49 something that impacted both uh mpls and

17:53 broadband and therefore there was no

17:54 good connectivity for the session smart

17:56 router to use and and you know my user

17:59 here um felt uh felt the impacts

18:03 um

18:05 so

18:06 you know to to summarize this this

18:09 really is ai driven sd-wan in action um

18:12 you know to kind of recap a little bit

18:14 on what we saw uh we saw our

18:18 uh

18:19 our day zero design of a wan um using

18:22 powerful templates that can flawlessly

18:24 be repeated across hundreds and

18:26 thousands of branches on day one we

18:29 brought our li our site live with

18:31 frictionless deployment and ztp from

18:33 freshly unboxed to secure wired wireless

18:36 and lan connectivity

18:38 and then on day two operations we saw

18:40 how our user experiences in the land

18:42 were assured to be the absolute best

18:45 they can be

18:46 and when they couldn't

18:48 uh marviss gave us

18:50 uh actionable information

18:52 on how we could improve it so

18:55 um you know sd-wans have been around a

18:57 while but uh um

19:00 you know many of them have shown to

19:02 deliver marginal gains and user user

19:04 experiences but for many it is at the

19:07 expense of operational hardship and

19:09 complexity

19:11 making sd-wan easy to deploy

19:14 design troubleshoot maintain you know

19:17 that's that's the nut that most haven't

19:19 cracked at this point and and you know

19:21 juniper's sd-wan ai driven sd-wan

19:24 as you saw

19:26 provides exceptional experiences not

19:28 only for end users on the land but for

19:30 the it operator administrator themselves

19:33 so how do you guys measure the health

19:35 and looking at the health analytics of

19:36 how you guys are measuring the underlays

19:38 and overlays you talk about things like

19:39 switching between mpls and internet but

19:41 how do you

19:42 measure the health of the global

19:44 internet i work a lot on the service

19:45 provider side and the health of the

19:47 internet you know and the transit and

19:49 peering that you're using the commodity

19:51 dia that you're using is a real

19:53 challenge how do you

19:55 tackle that problem when you're seeing

19:56 like you know global internet disruption

19:58 that we've seen more of to you know to

20:00 be honest in the last two years

20:02 yeah that's a great question and and

20:04 certainly the answer is going to be an

20:06 evolving one because um you know one

20:09 nice part about this being the cloud and

20:11 these sles um

20:13 the you know how exactly the state of

20:16 the art about how exactly we do that

20:18 will continue to

20:19 evolve and the efficacy of this will get

20:21 better and better

20:23 today

20:24 uh

20:25 you know like we mentioned every single

20:27 flow in the sd-wan ultimately on a

20:30 flow-by-flow basis uh is being monitored

20:33 for um a number of things so we look at

20:36 things like um

20:38 you know metrics like time to first

20:40 packet like say there's a tcp session

20:43 and we measure the time from the tcp

20:45 send to the time that that we actually

20:48 start to get um payload uh good put

20:51 running through that session um and that

20:54 becomes a metric that we feed up in into

20:57 the uh algorithms of of mist

21:00 we we measure when there's tcp resets um

21:04 and keep a running metric of that that

21:06 feeds up into the models um

21:09 uh when there's udp flows we measure

21:12 size of those flows you know

21:14 in fact the the teams sessions that you

21:16 see there um you know those are actually

21:19 detected based on trained models of what

21:22 um teams flows look like and so you know

21:25 it can tell when

21:26 there's a teams session that that you

21:29 know that is anomalous for some

21:31 particular reason so um you know at the

21:33 moment it's there's there's a heavy dose

21:35 of of um kind of flow level

21:38 telemetry that's that's feeding into

21:40 that to kind of understand what's going

21:42 on further out behind

21:45 but certainly in the future you know we

21:48 we we have the ability to look at probe

21:50 data and and other things as well first

21:53 thing uh we're not trying to solve the

21:55 solve the world's internet what we're

21:57 trying to get you to is mean time to

21:59 innocence right so you know because

22:02 every problem starts with the wi-fi

22:04 sucks or my network sucks until proven

22:06 otherwise so the first thing we're doing

22:09 is is it the wi-fi is it the lan

22:12 is it the my sd band eliminate those

22:15 things first and once you say hey it's

22:18 off-prem to me

22:20 then you look at the rest of the world

22:22 and so um the thing that we've been able

22:25 to do again we're not here to solve the

22:27 world's internet problems uh but they

22:29 impact our experience is first to make

22:32 sure that the iet team can say

22:35 confidently draw a line in the sand and

22:37 say this is on our side or not on our

22:40 site and then you say hey if it's not on

22:43 our side is it my next hop with lte or

22:47 5g we've just announced a partnership

22:49 with cradlepoint as an example

22:52 you know because they happen to be a lot

22:54 of the 5g routers in a lot of these

22:56 branches you know so so there's

22:58 interesting possibilities on how we

23:00 stitch the story together but our first

23:02 focus is truly mean time to innocence

23:05 for your it team and then we look at

23:08 what's extraneous to the network does

23:10 that make sense yeah no that's a totally

23:12 fair answer i get why you would focus on

23:13 on that perspective i just i think you

23:15 know i work in consulting now and mostly

23:18 on the service provider side but still

23:19 but i did work as you know the network

23:20 architect for a publicly traded company

23:22 and i know when i was in that position

23:24 in the meantime an instance is great but

23:26 the end of the day you still have the

23:27 problem the company's still down so you

23:29 know having that global health data of

23:31 you know if one of the one of the top

23:32 five you know transit carriers suddenly

23:35 has a bad day and i have 300 sites and

23:37 100 of them go down i don't know why you

23:39 know finding a way to get those

23:41 analytics into mist would be another

23:42 huge step in the evolution of this i

23:45 think and i don't know if anybody's

23:46 doing that because i haven't looked too

23:48 deeply into it but that was the thing

23:49 that stands out to me is you guys have

23:51 some great analytics for you know

23:53 figuring out what is within your control

23:55 but having that ai capability to me that

23:57 seems like a really logical extension of

23:59 the ability to measure that because you

24:01 know you're right like you're you're

24:03 trying is what is in my my team's

24:05 ability to control

24:06 but i also i still have to solve the

24:08 problem so that was basically the origin

24:10 of my question yeah yeah and so we've

24:12 just filed a patent on

24:14 solving the second half of the problem

24:16 so not for this uh nfd but i think at

24:19 mfd in probably about six months i want

24:21 to share with you what we're doing to

24:24 the rem the rest of the network uh um

24:27 definitely there's lots of very

24:28 interesting ai things we can do

24:30 for the rest of the network as well so

24:33 so hang tight

Show more