Transition to Cloud-Native and Distributed Edge Architectures in Service Provider Networks

0:00 [Music]

0:03 get connected

0:06 you're listening to get connected

0:09 thanks for joining us

0:13 in this series we discuss the various

0:16 issues shaping your industry how the

0:18 changes in supplier ecosystems impact

0:20 the way you work on a daily basis how

0:23 your customers consume the services you

0:25 create and how the industry as a whole

0:27 steps up to the demands of the digital

0:29 marketplace let's get connected

0:36 hello listener let's get connected

0:41 my name is chris lewis

0:43 by day i'm an independent industry

0:45 analyst

0:46 by night as i'm sure you know by now i'm

0:49 a podcaster

0:53 welcome to get connected where we delve

0:56 into the issues shaping the networking

0:58 industry and the experiences we all

1:01 benefit from by being better connected

1:05 in this episode we're continuing the

1:06 journey into cloud the transition to

1:09 cloud native and the shift to this more

1:11 distributed architecture for the service

1:13 provider environment

1:15 i'm delighted to welcome pavan kurapati

1:18 chief architect sp global architecture

1:21 at juniper back to get connected welcome

1:24 back pavan and please remind us of your

1:26 background and your role within juniper

1:29 thank you chris this is uh indeed my

1:31 pleasure to talk to you again i am the

1:33 chief architect of global sp

1:35 architecture for juniper and in my role

1:38 me and my team are responsible for all

1:40 the emerging technologies that our

1:43 service provider customers are investing

1:45 in and we make sure we develop the

1:47 solutions that are more forward-looking

1:50 for the sp customers

1:51 so to start us off then we're talking

1:54 about this continued journey into the

1:56 cloud and the migration the move the

1:58 change in architectures what are the

2:00 high level transitions that you're

2:01 seeing that are shaping the marketplace

2:03 for you so the biggest trends in 5g is

2:06 the transition to virtualization and

2:08 cloud native network functions let me

2:11 start with the ran virtualizing of ran

2:13 and running on general purpose hardware

2:16 instead of the appliances that are more

2:18 standard until 5g came so that's the

2:21 biggest trend i'm seeing in the ran you

2:23 know virtualizing of ran is going to

2:25 take place irrespective of whether you

2:27 talk about the orion ecosystem or not

2:30 desegregation through open ran ecosystem

2:33 is again taking off i was in barcelona

2:36 you know mobile world congress where

2:37 everybody is talking about how

2:39 virtualization of ran is going to save a

2:42 ton of cost for service providers

2:44 the sa core the standalone core in 5g

2:47 that's designed to be fully cloud native

2:50 and based on the cloud native principles

2:52 along with the services-based

2:54 architecture so that's going to change

2:56 how 5g is deployed kubernetes we were

2:59 talking about a lot of cloud native

3:01 principles here so kubernetes is going

3:04 to be the de facto orchestration

3:06 platform whether it is for the ran or

3:08 for the core

3:09 and in order to deploy all these things

3:12 before 5g

3:14 most of the mobility traffic used to be

3:16 backhaul to a centralized data center

3:19 where you would bring all the traffic to

3:20 a central location and from there you

3:22 would serve but that is changing quite a

3:25 bit now so telco cloud is transitioning

3:28 from centralized to distributed edge

3:30 architecture to be able to host these

3:32 cloud native network functions that i

3:34 just talked about the ran and the core

3:37 so i think in my opinion virtualization

3:40 of ran the 5g standalone core kubernetes

3:43 becoming a de facto orchestration

3:45 platform and

3:46 moving from centralized to distributed

3:48 edge architecture these are the key

3:50 trends that we are tracking and working

3:52 actively on i totally agree and the fact

3:54 is that we're talking about this more

3:56 distributed architecture where we're

3:58 moving the resources to where it's

4:00 required to actually execute right you

4:02 mentioned rand virtualization

4:04 specifically in there at the cell site

4:07 you know juniper is known for its work

4:09 around the transport as you said and the

4:11 traditional routing that we've seen in

4:12 place how are you seeing the cloud

4:15 native

4:16 developing and how is run virtualization

4:18 impacting that

4:20 that's a great question chris so

4:22 the ran as i said is getting virtualized

4:25 and especially with iran desegregation

4:28 there are cases where you run the dus at

4:30 the cell site itself and if it's a

4:32 virtualized u it runs on a general

4:34 purpose x86 based server

4:37 so we do have physical sell side routers

4:40 juniper provides these physical sell

4:42 side routers for front hall and mid-hall

4:44 connectivity and two main important

4:47 aspects of sell side routers are timing

4:49 and synchronization especially

4:51 classy timing and hardened cell side

4:54 routers as well as the routing stack

4:56 required

4:57 to support the network slicing

4:58 capabilities

5:00 now for some distributed ran

5:01 architectures especially where you know

5:04 you only require sub 25 gig or sub 50

5:07 gig mid hall capacity this is where

5:10 things become lot more interesting

5:12 because here you might save operational

5:15 expenses if you eliminate an additional

5:17 physical device at the sell side and

5:20 that's where juniper is working on what

5:22 we call as a cloud native router

5:24 the idea here is to take best of both

5:27 worlds which is junos routing that runs

5:30 on all our routers

5:32 we take that and we containerize that

5:34 which is our containerized routing stack

5:36 as a control plane and then we take our

5:39 contrail dpdk forwarding plane and

5:41 deploy it as two different parts which

5:44 forms the brain of junos and the

5:47 forwarding plane of contrail and that's

5:49 integrated with the kubernetes cni

5:51 framework that i talked about before so

5:55 that cloud native router is where we

5:57 think the dram deployments with the sub

6:01 50gb hall will go and it's so

6:04 interesting because we're bringing so

6:05 many things into into the mix here so

6:07 cloud native routing does sound very

6:09 interesting at the sale side but what

6:11 kind of integration are you looking for

6:13 and expecting from the ran and the

6:14 hardware vendors oh great question again

6:17 very minimal integration from the rand

6:19 vendors in fact the networking for the

6:22 ran especially when you're looking at

6:24 the du and let's say you require three

6:27 slices one say going to vpn green

6:31 another to blue another to red going to

6:33 three different locations

6:35 typically you would tend to put all that

6:37 information in kubernetes manifest and

6:40 we automatically translate that into

6:42 junos construct without you doing

6:45 anything as a rand vendor you don't have

6:47 to do any additional configurations for

6:50 example to make that happen

6:52 from hardware again we understand that

6:54 this is a resource constrained

6:56 environment we are talking about having

6:58 very less cores for routing

7:01 and we do understand that and we operate

7:03 in a very limited core budget so for

7:05 example today we only need about two

7:07 cores to run our entire routing stack

7:10 and still be able to provide the

7:12 required mid hall capacity

7:14 in addition to that we are also

7:15 exploring the offloads for example on uh

7:18 offloading the entire data plane on

7:20 fpgas as well as on the smart mix

7:23 what i always like about this

7:24 development in the industry is that

7:26 nobody can do everything but the skills

7:29 being brought to the table by everybody

7:31 really complement each other it's a

7:33 convergence of of it of transport of

7:36 networking of all of these things but i

7:38 think what really comes to mind when we

7:40 talk about this is that the scale of

7:41 these virtual elements is going to be

7:43 absolutely enormous

7:45 how are you going to automate the

7:46 install and the provisioning of all of

7:48 these pieces oh you are absolutely right

7:51 as we talked about before we are talking

7:53 about tens of thousands of cell sites

7:55 that will require this kind of

7:58 features and functionality so yes the

8:00 scale is enormous so again the key idea

8:04 here is to follow kubernetes best

8:06 practices because kubernetes has

8:08 built-in aspects that will take care of

8:10 the life cycle management just to give

8:12 you some examples so we provide helm

8:15 charts that will define and you know

8:17 install

8:18 all the routing stack the parts that are

8:20 required so we use helm charts it's

8:22 literally just a one line that will

8:24 bring up the entire parts in terms of

8:27 provisioning i talked to you about that

8:29 before we can take

8:31 the networking constructs from the

8:33 kubernetes manifest and we automatically

8:35 translate that terraform is very popular

8:38 these days junos supports terraform

8:41 providers so you can actually come in

8:43 and provision things using terraform in

8:46 addition to that it is a regular junos

8:49 it's just in a cloud native form so we

8:51 support all the interfaces that are

8:53 exposed in junos for example

8:55 either netconf and you can come through

8:58 psap you can talk bgp you can use our

9:01 rpd apis you can use exactly the same

9:04 way that you would use any other juniper

9:06 device so that's really the key for us

9:09 automation is very very important at

9:12 this scale and we are doing everything

9:14 to utilize kubernetes ecosystem you're

9:17 talking tens of thousands potentially

9:19 even more than that the densification of

9:21 the network has been moving to the 5g

9:23 world and farther on into the 6g world

9:25 means that we're going to get denser and

9:26 denser and denser so forward-looking we

9:28 really have to nail this in terms of

9:30 making sure we can automate and deliver

9:32 it right right exactly absolutely so

9:34 that looks to be from a desegregation

9:36 perspective a great story in the

9:38 transport

9:39 but talking about disaggregation more

9:41 generally and especially in the ran

9:43 how do you see the open ran and how does

9:45 juniper see the open run environment and

9:47 the benefits it brings to the service

9:49 providers right as i said i was in

9:51 mobile world congress and i could hear a

9:54 lot of people being very positive about

9:57 oran

9:58 taking off especially the open

10:00 interfaces that are essential

10:02 for the oren ecosystem right so

10:04 disaggregation

10:05 traditionally when you talk about tran

10:07 it's vertically integrated into an

10:09 appliance

10:10 now the disaggregation allows you to

10:13 divide it into multiple components and

10:15 you can mix and match right you can mix

10:17 and match the ru's du's and cu's with

10:20 multiple vendor ecosystem so that's

10:22 really really healthy for the service

10:25 provider networks

10:26 one of the key aspects we are also

10:28 noticing is you know the decoupling of

10:30 ran intelligence and removing some part

10:33 of the intelligence away from the

10:35 traditional appliance and deploying it

10:37 as applications like use aiml to get the

10:41 data from the ran and deploy it as an

10:43 application ecosystem so we see that

10:46 rick which is the ran intelligence

10:48 controller

10:49 that provides a thriving application

10:51 ecosystem in the form of x apps and r

10:53 apps in the marketplace so all the

10:56 intelligence is going to be i think in

10:58 the applications we're talking about

11:01 spectrum utilization efficiencies for

11:04 example like 30 to 40 percent of savings

11:06 in spectrum utilization sustainability

11:09 as an aspect sla management and so many

11:12 applications that can be developed in

11:15 this environment especially in the rand

11:17 desegregation so we

11:19 do believe oran especially in the ric

11:21 and the application ecosystem

11:23 that

11:24 is provided through the oran approach is

11:27 going to be

11:28 crucial for 5g deployments

11:31 that supply ecosystem is so important

11:33 but speaking specifically about rick

11:35 which is getting such a lot of attention

11:36 at the moment how do you see that being

11:38 deployed we are designing rick to be

11:41 cloud native from day one the platform

11:43 itself is being built on cloud native

11:45 principles so what it means is rick can

11:47 run on service provider private cloud

11:50 platform that is there in their on-prem

11:53 or if they're using a hyperscaler

11:55 platform in their premises the rick can

11:57 run on that platform as well that

12:00 flexibility of where the ran is running

12:02 obviously fits in with what we're seeing

12:04 in the server provider community where

12:06 they're looking at all sorts of

12:07 different implementations their own

12:09 environments as you said or indeed in

12:10 the hyperscaler world

12:12 now we've talked a lot about ram

12:15 in mobile networks the epc in the 4g

12:18 world is where virtualization really

12:20 started

12:21 the 5g standalone core is cloud and

12:23 microservices based what opportunities

12:26 does juniper see in the virtualization

12:28 of the 5g sa core great question chris

12:31 so 5g core vendors as i talked to you

12:33 about

12:34 5g core previously it is designed to be

12:37 services based and you will require an

12:40 orchestration platform to host these

12:42 network functions and typically 5g core

12:44 vendors try to sell their own vertically

12:46 integrated platforms to host their core

12:49 applications claiming that you will get

12:51 better performance of their core running

12:54 on their platform but i think we are

12:56 seeing operators looking at this as an

12:59 opportunity to disaggregate and

13:01 build a horizontal cloud platform where

13:04 they can run these 5g core network

13:07 functions to avoid the vendor lock-in

13:09 and this is where we see a big

13:11 opportunity for contrail so today

13:15 we have

13:16 5g core even the 4g virtualized core

13:19 running on contrail based horizontal

13:21 cloud platforms where contrail provides

13:24 the overlay networking initially we did

13:27 all this work with openstack but now we

13:29 are seeing this migrate to kubernetes

13:31 based ecosystem so from our point of

13:34 view we think contrail networking

13:37 provides huge differentiation especially

13:40 with respect to providing advanced

13:42 networking features such as you know bgp

13:44 as a service advanced security

13:47 especially in the multi-tenant and the

13:49 name space network isolation and other

13:51 aspects the multi-cluster networking

13:54 which is very important in the 5g

13:56 standalone code deployment so all of

13:58 these are key differentiators from

14:01 networking point of view when you are

14:02 talking about building these horizontal

14:04 platforms and that's where we think

14:06 contrail networking is evolving into

14:09 more cloud native

14:10 form we call it cloud native control

14:12 networking or cn2 or cn square

14:15 the other aspect which is also very

14:17 important is the underlay networking

14:20 where you have leaf and spine you need

14:22 to provision and configure but more

14:24 importantly it need to integrate well

14:27 with the orchestration platforms that

14:29 you are using

14:30 for the standalone core and here is

14:32 where we think our intent based fabric

14:35 management which is abstract that

14:37 juniper acquired plays a very important

14:40 role

14:40 and we have been helping various

14:42 operators with the appstra as well as

14:45 contrail based horizontal cloud

14:48 platforms

14:49 to be able to host these 5g cloud native

14:51 applications especially in the

14:53 standalone core what is fascinating here

14:55 is as you quietly write pointed out the

14:57 underlay and the overlay building these

14:58 things together and you brought those

15:00 into juniper to help in the in your

15:02 delivery but what we're also seeing is

15:05 announcements about some of the telcos

15:06 working with the hyperscalers to move

15:08 workloads such as 5g core into those

15:11 hyperscale environments so we've seen it

15:13 with dish with aws and we've seen it

15:14 with att with azure how do you see

15:18 juniper fitting into this this

15:20 hyperscaler ecosystem

15:22 again another great question we do see

15:24 these strengths as you said moving into

15:27 hyperscalers especially running these

15:29 virtualized network functions in the 5g

15:32 space whether it is ran or core requires

15:35 you to invest in the infrastructure and

15:37 the platform so one of the aspects that

15:39 some operators feel

15:41 is easier to do is simply run it in the

15:43 hyperscaler environment right where

15:46 hyperscalers would provide the required

15:48 infrastructure and platform and you can

15:50 just take care of the network functions

15:52 or the applications here whether it is

15:54 the core or the ran

15:56 i think some of the challenges that we

15:58 are seeing is

16:00 it's not just about having the 5g core

16:02 running in a hyperscaler environment

16:04 things just come up but it's about

16:07 connecting these 5g core network

16:09 functions or interconnecting them

16:12 that's where there are a lot of

16:14 challenges because the networking

16:16 requirements for example to host the

16:18 standalone core you will need network

16:20 segmentation you will need to extend

16:23 your current vpns in your on-prem data

16:26 centers into the hyperscaler data

16:28 centers so terminating those overlay

16:32 tunnels as well as to be able to provide

16:34 network segmentation these are

16:36 challenges that you will find in all

16:38 three hyperscalers to be honest and

16:40 that's where we come into picture we are

16:43 very actively working with all the three

16:45 hyperscalers to address these networking

16:48 requirements i talked about juniper

16:50 cloud native router in the oran space

16:53 before it's actually the same cloud

16:56 native router that we are

16:58 working with the hyperscalers to be able

17:00 to deploy in an auto scale cluster which

17:04 will provide these network segmentation

17:06 capabilities

17:07 and be able to

17:09 interconnect these vpcs that are hosting

17:13 the 5g core elements so in a nutshell we

17:16 think that is an opportunity for us

17:19 where if the service providers are

17:21 looking at hyperscalers to host these

17:23 network functions we come in and help

17:25 them in connectivity and automate all

17:28 the connectivity aspects whether it is

17:31 between their edge cloud and public

17:33 cloud or between the vpcs within the

17:35 public cloud

17:36 and it's fascinating isn't it it's these

17:38 little elements in between so it's

17:40 working with the hyperscalers so it's

17:42 not one part of the ecosystem imposing a

17:44 way of working on the other is the

17:46 visibility that you're bringing to the

17:47 table which will allow all of these

17:49 different flows to work correctly in the

17:51 long run absolutely absolutely yeah and

17:54 we've talked primarily here about the

17:56 wireless environment because of course

17:57 5g gets a lot of that attention from the

17:59 marketplace but what about the y line

18:02 side do you see similar trends on the

18:04 wireline side of the the argument oh

18:06 yeah i mean the desegregation that is

18:08 true even in the wireline world take an

18:10 example of bng we are talking about cups

18:14 which is the control and user plane

18:15 separation that's already happening so

18:18 the bng control plane is already

18:20 separated and it can be run as micro

18:23 services in any cloud platform so today

18:27 we have our bng control plane which is

18:29 completely micro services based it can

18:32 run on on-prem cloud um if you are using

18:35 any on-prem hyperscaler infrastructure

18:38 like aws wavelength for example we can

18:41 run the bngc even on hyperscaler

18:44 infrastructure the key is you need to

18:46 design it right follow the kubernetes

18:48 based ecosystem and the same principles

18:51 apply to wire line world as well

18:53 it's great to hear because i think all

18:54 too often we don't think about the fact

18:56 that as consumers we take wi-line

18:58 services we take wireless services we

19:00 want that quality of service to be

19:02 delivered throughout and what is

19:04 becoming increasingly evident is that we

19:06 need the different players in the

19:07 ecosystem so people like yourselves

19:09 people like the hype scalers and of

19:10 course the service providers themselves

19:12 to be working much more closely together

19:14 to have that management to make sure

19:16 that reliability gets delivered through

19:18 to all of us as users

19:20 pavan

19:21 fantastic to have you with us again

19:22 thank you so much for having joined us

19:24 on get connected it was great talking to

19:27 you chris uh thank you so much for

19:28 having me

19:30 so there you have it listener we

19:32 continue this journey into cloud we

19:34 continue bringing together the elements

19:36 of the cloud environment the service

19:38 providers and the ecosystem of suppliers

19:41 to be able to work with everybody to

19:43 deliver the quality of service that

19:45 underpins the future of the digital

19:47 marketplace

19:48 not only is it required in the wireless

19:50 world it's required in the fixed world

19:51 as well we bring together the benefits

19:53 of having this continuous integration

19:56 the continuous management of all of the

19:57 services we're beginning to build that

19:59 momentum as cloud brings all of these

20:01 capabilities to give the scale that we

20:04 need to support all of those services in

20:06 the future

20:09 thank you for joining us on get

20:10 connected thank you to pavan thank you

20:13 to juniper and thank you too for

20:15 listening

20:18 you've been listening to get connected

20:21 you've heard from us now we'd love to

20:23 hear from you

20:24 tell us what you think via twitter using

20:27 the handle at juniper networks

20:30 and if you like what you've heard

20:31 remember to tell your colleagues and

20:33 friends

20:34 thanks for listening to get connected