Tech Bytes

Customer Case Study: Advania Chooses Juniper Apstra for Simplified Data Center Operations

Industry Voices Data Center
Title slide with two people working in a data center on the left. Text on the right says, “Multi-vendor support. Freedom of choice and avoid vendor lock-in. Multiple data centers can be connected seamlessly.”

How Advania made its data center so much simpler.

In this episode of Tech Bytes, hear from Darko Petrovic of Advania, one of Iceland’s largest IT solutions providers, on how his company uses Apstra to operate its data centers, simplify its operations and improve user-experience.

Show more

You’ll learn

  • Why Advania chose the Apstra system over other competitors  

  • How Apstra is easy to use and, according to Petrovic, “effortless”

  • The many benefits Advania realized through Apstra automation, including continuous validation and multivendor support 

Who is this for?

Network Professionals Business Leaders

Host

Tech Bytes

Guest speakers

Darko Petrovic
Principal Engineer, Advania (Iceland)

Transcript

00:00 today on the techbytes podcast sponsored

00:02 by juniper we talk with a customer of

00:04 juniper's abstra intent based networking

00:06 data center software our guest is darko

00:08 petrovic he is principal engineer at

00:10 advania iceland avani is an i.t

00:12 solutions company offering managed and

00:13 professional services and i.t

00:15 infrastructure and integration uh darko

00:17 welcome to the podcast so what drew your

00:19 interest to appstore's product

00:21 first off like you said that vanier is

00:24 working for advania managed service

00:26 provider we are one of the biggest if

00:29 not the biggest in

00:30 msp in iceland

00:33 our main goal going with abstract was we

00:36 wanted to automate our data center as

00:39 much as possible and go away from the

00:41 traditional layer twos that many data

00:44 centers are having trouble

00:46 in the past years

00:47 so we were looking at different

00:49 solutions vendors um uh even thinking

00:53 about doing ourselves with some kind of

00:56 habitation with ansible

00:58 okay so you're looking for an automation

01:00 solution you maybe investigated a little

01:03 bit of doing it yourself but what was

01:04 there something that about appstore that

01:06 you thought oh wait this is going to go

01:08 better for us so

01:09 the first

01:11 impressions from abstract i had a couple

01:13 of years ago when i was doing my

01:15 investigation about uh

01:17 similar solutions that are out there but

01:19 it it went

01:21 uh to the to the side that abstract is

01:24 way more than and offering way more than

01:27 other solutions and vendors there

01:30 just because

01:32 the level of automation and the level

01:35 that

01:36 it's doing the provisioning and the

01:38 whole networking bits and pieces that

01:41 are under the bonnet it's

01:43 so easy and it's effortless basically

01:46 effortless so what you've found is that

01:49 putting abstract into your managed

01:51 services network into the underlay of

01:53 that data center environment has really

01:55 just reduced the friction about making

01:57 changes right not even that the margin

02:00 of human error is minimal so

02:04 how the abstract does that is

02:06 it has basically every in each scenario

02:09 that you can have in your data center

02:12 and based on your inputs he's going to

02:14 calculate the option we're going to

02:16 calculate is that feasible or not if

02:18 it's visible it's going to apply it on

02:21 your network

02:22 it's like pre-applying the configuration

02:25 the app calculates is the the links are

02:28 okay is the routing okay and based on

02:31 all that pre-calculated stuff it expects

02:34 some results when you commit that

02:36 changes so the human

02:39 when it's the engineer when it's uh

02:41 provisioning vlans vague slangs or links

02:45 servers new servers it's basically four

02:48 clicks to provision that so what you're

02:50 saying is actually sanity checking if

02:52 you're trying to make a configuration i

02:53 want this vlan here to be connected to

02:56 this vlan but it also needs to connect

02:58 to this

02:59 segment of the network it's actually

03:01 going to sanity check that everything

03:03 that you want or you think you want to

03:05 configure is actually possible

03:07 correct even the ips the vlans it's

03:10 going to cross-check the vagus lines the

03:12 v-lines it's going to complain if you're

03:14 using the same vlan

03:16 or the same ip addresses ranges or

03:18 scenes right correct that's that's

03:20 really cool what can i ask just for a

03:22 second here what because it's obviously

03:25 you as a managed service provider you've

03:26 got a choice of switches and you're

03:27 probably operating at a reasonable sort

03:29 of scale

03:30 what sort of switches physical switches

03:32 did you go with so

03:34 one of the huge elements that we wanted

03:37 from our fabric is that we don't want to

03:41 do

03:42 vendor lock that's especially important

03:44 for the larger data centers

03:46 when they have multiple vendors inside

03:49 their network and juggling between the

03:51 features you know

03:54 so we wanted something that's going to

03:56 give us the freedom of choice in the

03:59 future years if we don't like or we are

04:02 going to have a bad relationship at that

04:04 moment with someone wender

04:06 all right we have an alternative so it's

04:09 giving us leverage basically in the

04:12 negotiations for the underlay hardware

04:15 at this point we went for the dell

04:18 and where and with their enterprise

04:20 sonic

04:21 distribution

04:22 sonic is based on the fr so like the

04:26 majority of vendors are i know the

04:29 vmware also uses the facebook

04:32 um so yeah

04:34 the the the sonic is so del owen

04:37 dell's open networking switches with the

04:39 sonic

04:40 uh distribution that they've got which

04:42 is supported right dell supports that

04:45 and then you're running abstract on top

04:46 of that so you correct you've actually

04:48 stopped not taps on top abstraction just

04:51 let's say a controller you know yeah so

04:54 the the operating system is sonic

04:57 the the dell itself is shipping so that

05:00 has two flavors of of operating systems

05:03 they have their stock operating system

05:05 os 10 or nine

05:07 um and then there is an enterprise grid

05:10 for their operating system that's sonic

05:12 okay so you've got dell sonic underneath

05:14 and you've got some flexibility there

05:16 because you could either replace the

05:17 dell hardware with another open switch

05:18 running a sonic you could replace the

05:20 sonic with another os but more

05:22 importantly abstra

05:23 over the top here gives you the freedom

05:25 to change the hardware underneath

05:27 i guess that the idea here is that

05:29 appstre is doing the provisioning of the

05:30 switches for you you're not

05:33 using the command line or excel

05:34 spreadsheets or text files to configure

05:36 them right

05:37 nothing nothing so not it's not just

05:40 giving you freedom to change the

05:42 hardware it's you can keep the same

05:44 topology same ip same everything so

05:48 everything is the same from your

05:49 perspective but the hardware you can

05:51 swap

05:52 of course it's not that easy in uh in uh

05:55 in the real world in sense you need uh

05:58 good planning and migration scenario and

06:00 everything but

06:01 at the bottom line yeah

06:03 it's everything that you have everything

06:05 that's running right now

06:07 you can change and keep the same

06:10 topology same layer 3 configurations

06:13 when you finish migrating to another

06:15 vendor okay so i think that's something

06:16 key to talk about is that what you're

06:18 getting with abstra is not just this

06:21 uh sanity check this help with

06:23 configuration this automation lever

06:25 layer but you're also essentially

06:26 ensuring that you've got uh you can run

06:29 a multi-vendor network

06:31 that's correct how are you multiple data

06:33 centers can be connected seamlessly

06:36 really seamlessly i mean i'm telling

06:38 this from my experience that i had

06:41 before with other vendors i was working

06:44 before advani i was working for a global

06:47 integrator so i went through a lot of

06:50 different products based on the for the

06:52 sdn solutions in the data centers and

06:55 appstra is

06:57 maybe the best choice

06:59 if you want to go

07:01 having something nice easy

07:04 uh and managing many manageable

07:07 so it actually works because one of the

07:09 stories i hear a lot is with sdn

07:11 controllers is that people spend an

07:13 awful lot of time keeping the controller

07:15 running or installed or maintained

07:17 and they actually don't have a lot of

07:19 time left over to do other work because

07:21 you're also running an nsx you're

07:23 running a vmware nsx deployment for the

07:25 second part of the networking is there

07:27 an integration between the two what's

07:29 that like with nsx and apps working

07:30 together

07:32 so our current

07:35 network is compromised of three

07:38 different

07:39 key components one of them is nsxt the

07:43 other one is the data center fabric

07:45 that's running on apps and then we have

07:47 that traditional mpls core for

07:51 interconnection with other carriers and

07:53 customers

07:54 so the thing that's appsha giving us uh

07:58 in in the middle is connectivity between

08:01 the data centers

08:03 we're not touching the mpls core in that

08:05 perspective so we can stretch a vlan or

08:07 l3 even through multiple data centers so

08:11 anycast gateway is also there and

08:15 it doesn't even hit the core so the core

08:18 is there just a transport network no

08:20 configuration there

08:22 from the configuration perspective of

08:24 the nsxt so appsha is doing read-only

08:29 queries to the nsxt manager where can

08:32 you where you can see all the vms the

08:34 vlans the

08:36 the vds's the the the

08:38 the port groups and based on that you

08:40 can map out your abstract switches

08:43 vlans uh the policies or whatever you

08:46 want so you what you're actually saying

08:48 there is that you're actually

08:49 integrating the nsx and the abstract

08:50 together it's not a it's not a a a read

08:54 write and everything's coming together

08:56 and you know mystical magic happens it's

08:58 abstract looks at the nsx configuration

09:00 and knows what's happening in the

09:01 overlay and can work to help you with

09:03 that is that right

09:05 correct

09:06 well it gives you information

09:08 it doesn't do the configuration actual

09:10 configuration of the nsxd that's done

09:13 separately right but it's done in nsx

09:16 right yeah right yeah but the everything

09:18 that you configured there on the net on

09:20 the on the the segments on the t ones

09:23 the t zero t one everything is visible

09:26 in yapstra i mean the the the vms to

09:29 which uh uh network adapter is connected

09:32 then you can map out that you can even

09:34 forbid traffic of course that's also

09:36 possible in the nsxt but additional

09:38 security is available for you on the

09:42 aperture level so you can right so you

09:44 can actually integrate the micro

09:45 segmentation policy that you've got from

09:47 nsx with what's in abstract it's not

09:49 that appstore is going to configure the

09:51 nsx but it's aware of what the nsx

09:53 configuration looks like and you can see

09:55 it

09:55 yeah and that means a lot to your

09:57 network guys i mean to see what's what

10:00 they're working with also the telemetry

10:02 of the abstract is fantastic it's a

10:06 the telemetry is way better than

10:08 anything that then you're going to use

10:10 on your nms

10:12 it's a go-to platform for for your

10:14 insights on your fabric you can see

10:17 there everything from the bandwidth to

10:19 the errors to the dropouts to the

10:22 notification to

10:23 even

10:25 even the the the the

10:27 microbursts are seen there

10:29 so what you're actually saying is your

10:31 whole underlay is run by abstract it's

10:33 monitoring the cable performance by

10:34 looking for drops and signal changes

10:37 on the optical because it actually

10:38 watches the optical state of the fiber

10:40 optic connections to see if they're

10:41 working same for copper slightly

10:43 different though depending on the type

10:45 of operating system and so forth but if

10:46 you want to know uh you know is my spine

10:48 running at capacity is there a hot is

10:50 there links

10:51 in the ecmps

10:54 you know leaf spine architecture that

10:55 are running too hot you can find out

10:57 it's all just part of apps you don't

10:58 have to go and find another system to

11:00 watch that

11:01 that's correct i don't know for copper

11:03 though because we don't have anything on

11:05 copper these days

11:07 but yeah for the fiber yeah and there is

11:10 like a really nice graphic

11:12 on appshop for your link you know for

11:14 this for the spinal leaf links and it's

11:17 as your throughput go as your bandwidth

11:19 goes high the that visualization

11:21 visualization of the that link goes

11:23 thicker you know

11:25 so

11:26 you have all the numbers they're

11:29 running but yeah when you see the visual

11:31 capacity what's your limit oh and that's

11:34 that's really nice and really helpful

11:36 the really yeah because you know where

11:38 you are you're like you're mean time to

11:39 innocence like it's not the network and

11:41 you can prove it because you can look at

11:42 everything and go like bandwidth fine

11:44 interfaces

11:46 it evens it even draws out your entire

11:48 topology how your current topology looks

11:51 like all the links are there between the

11:53 spines the leaves and the servers

11:55 everything is there like

11:57 real lines you know i didn't you didn't

11:59 have to draw it or configure it or yeah

12:02 so i mean the only thing that that's

12:04 needed is when you build out your fabric

12:08 so you're gonna click and on a lot of

12:11 stuff uh rename a lot of stuff so that

12:14 initial setup is is a little bit

12:17 nasty as in never so that's like

12:20 mandatory inevitably

12:22 each and every solution but

12:24 the nice thing is that

12:26 it's going the abstract we're going to

12:27 build out your entire fabric just based

12:30 on the lldp

12:31 so that's it you can

12:33 plug in your your spine and leaf links

12:36 to whatever port and don't even

12:39 put an interface number in abstract it's

12:42 going to do it by for you for you and

12:44 then it's going to draw you out how your

12:46 topology looks like

12:48 so doc i wanted to ask you know whenever

12:50 you bring in any kind of automation

12:51 solution that can there's a learning

12:53 curve it might disrupt the way people

12:55 are used to working was there any

12:56 friction with you in the team trying to

12:58 get used to abstract in this new way of

13:00 working well it was

13:02 on my part because when we started

13:04 testing this on puck it was version

13:06 three we were testing version three a

13:09 major abstract gave a lot of changes in

13:12 version four

13:14 so version four is you'll get a separate

13:17 uh a separate

13:20 sections for configuring connectivity

13:22 for external templates uh route maps et

13:25 cetera et cetera so in that's a sense i

13:28 me personally had some hey you were the

13:30 bottom

13:32 yeah yeah yeah yeah but for the rest of

13:35 the guys it was easy because they they

13:37 don't know anything

13:38 for the

13:40 commenting yes

13:41 they're coming in clean they they're not

13:43 thinking of all the configuration that's

13:44 going on under the hood and trying to

13:46 work out yeah so the learning curve is

13:48 not that steep it does have like i said

13:51 with every new solution that you

13:53 came across it does have any

13:56 a little bit

13:57 a hard time to get around it but with

14:00 this it's not that hard so i guess the

14:03 takeaway is that um

14:05 sort of the intent of my question was

14:07 like it sounds like you can get

14:09 a fairly good time to value a time to

14:11 where you're actually using it as

14:12 opposed to just learning it

14:14 like i said from my experience with

14:16 other vendors

14:17 and if you ask me now i would choose

14:20 appshop or any other vendor anytime it's

14:23 really proven that good for you compared

14:25 and you've got experience like you said

14:26 you came from an integrator

14:28 so you've actually had experiences on

14:29 other systems and you're pretty happy

14:30 with this one well the bottom line is

14:32 you don't need an engineer that's that

14:34 knows cisco juniper i don't know uh dell

14:38 uh huawei whatever you don't need that

14:40 kind of engineer right now in your data

14:42 center you need that not for everyday

14:44 work you only need it when you're making

14:46 critical changes yes yeah so you don't

14:49 you you just need a guide that knows to

14:51 click with a mouse

14:52 [Laughter]

14:54 i mean it's trivial comparison but as in

14:57 a nutshell it's like that basically i

15:00 mean when you're running a managed

15:01 service provider like advania is you

15:03 don't want to have high quality

15:06 highly trained high priced engineers

15:08 doing everyday work

15:10 you want to have them doing deep work

15:12 like

15:13 firewall deployments and rollouts and

15:15 reviews of structural fail like root

15:18 cause analysis and that type of stuff

15:20 and as a senior when i was a senior

15:22 engineer i spent way too much time

15:24 doing sculpture config yeah or just

15:26 configuring vlans because nobody else

15:28 could be trusted to do it and it wasn't

15:30 fun right

15:32 no it wasn't i mean like i said

15:34 absolutely is doing a sanity check for

15:36 you so

15:37 the human error is well minimal

15:41 it it really is

15:43 so

15:43 you don't have to i mean you have to

15:45 worry about when you get some errors on

15:47 the fabric itself and then you're going

15:50 to need one engineer or senior engineer

15:52 to t-shoot that but apart from that yeah

15:55 it's everyday provisioning and that's it

15:57 you just put a good foundations in

16:01 in the in the fresco when you build it

16:04 when you're building your fabric and

16:06 just template it

16:08 all right well that does bring us to the

16:09 end of our conversation darko thanks for

16:11 joining us and if you've peaked people's

16:13 interest in app store and they want to

16:14 find out more they can go to

16:16 juniper.net

16:17 packet pushers slash appstra that's

16:20 juniper.net packet pushers dot abstra

16:22 we'll also have that link in the show

16:24 notes that accompany this podcast again

16:25 darko thanks for joining us and thanks

16:27 to juniper for being a sponsor if you

16:28 like this episode you can find it and

16:30 many more find free technical podcasts

16:32 and our community blog it's all at

16:33 packetpushers.net follow us on twitter

16:35 we're at pack of pushers find us on

16:37 linkedin rate us on apple podcast and

16:39 last but not least remember that too

16:41 much networking would never be enough

Show more