Architectural Shift in the 5G Era | Juniper Networks at Asia Tech Singapore CommunicAsia 2021

0:00 [Music]

0:11 hello and welcome

0:12 to this session on architectural shift

0:14 in 5g

0:15 my name is nitin wake and i'm joined

0:17 today by my colleague aditya khal

0:19 we're both solution architects in the

0:21 juniper networks apac team

0:23 and today we want to discuss with you

0:24 about how 5g is

0:26 impacting the operator networks and what

0:28 architectural shifts it's causing inside

0:29 the networks

0:30 so let's get right into it

0:34 so the 5g era has already begun we've

0:36 got about 158 different operators who've

0:39 launched services on 5g

0:40 100 million plus subscribers that are

0:42 already using id services

0:44 and about a trillion dollars of

0:45 cataracts that has been allocated

0:47 specifically for 5g

0:48 investments between 2018 to 2025.

0:51 if you look at from the perspective of

0:53 the segments that is going to impact

0:55 it includes enterprise government as

0:58 well as consumer services

1:00 but the key question that comes up quite

1:02 often is that what is

1:03 that killer application for 5g

1:06 unfortunately that's a big that's a

1:07 question that's difficult to answer

1:09 but if we look at some of the high level

1:11 analysis and trends it's very clear that

1:14 while we don't know what that killer

1:16 application is going to be we definitely

1:18 know that the killer segment

1:19 is going to be enterprise there's about

1:22 a 3.4 trillion dollars that's expected

1:24 to be invested

1:25 by the enterprise within this 5g

1:28 technology

1:29 and this is just the 5g mobile

1:31 technology

1:32 along with that there's going to be

1:34 cloud and ai and all the other

1:36 investments that will go

1:37 along with this so the enterprise is the

1:40 segment that is going to

1:41 be the largest investor in 5g as well as

1:44 is going to be the largest

1:45 benefit reaper out of this technology if

1:48 you look at some of these industries

1:49 such as industrial automation

1:51 industrial automation is expected to

1:53 have 18 of this 3.4 trillion

1:56 dollar share of the investment and this

1:58 is going to impact everything from

2:00 any machines or the smallest chip which

2:02 is going to be either tracked

2:04 or controlled through wireless

2:05 technology similarly if you look at

2:08 other

2:08 other areas like healthcare or robotics

2:11 there's going to be a huge

2:12 impact that 5g is going to have in these

2:15 domains

2:15 arvr for example is going to impact the

2:18 retail segment

2:19 in a very significant way of how we will

2:21 interact and shop

2:22 in the new edge retail malls if you look

2:26 at the consumer segment

2:27 the consumer segment the cloud gaming is

2:29 expected to be one of the biggest

2:31 revenue spinners for 5g and the whole

2:34 gaming

2:34 ecosystem is expected to change from the

2:36 handheld games towards

2:38 cloud gaming that is delivered through

2:40 the edge using 5g

2:42 similarly ar vr here vr is approximately

2:45 20

2:46 of the total traffic inside the korea

2:49 network where they've launched 5g

2:50 and this is very exciting in terms of

2:52 the new technologies that

2:54 5g is enabling fwa is one of the

2:57 first applications that we will see on

2:59 the residential side for 5g

3:02 from a government perspective the focus

3:04 is going to be on energies and utilities

3:06 because

3:06 that's where a lot of the

3:08 transformations are expected to happen

3:10 and also in segments like smart cities

3:12 and automotive where there's going to be

3:13 a lot of public private partnership

3:15 in terms of how we deliver these

3:17 services while we don't know what that

3:19 killer application is going to be

3:21 we definitely know that 5g is going to

3:23 set that base on top of which

3:25 we will be able to you know first of all

3:27 launch these applications that we

3:29 already know but also be prepared for

3:30 those unknown applications that are

3:32 expected to come in the future

3:34 so with all of this said one of the

3:37 questions that again

3:38 comes up is that what is it about 5g

3:42 that makes it different

3:43 is it just about the speeds or is there

3:45 something more to it

3:47 so let's look at the two of the

3:49 components of 5g that

3:51 are critical to to create this

3:53 differentiation

3:54 for 5g and that helps it to enable all

3:56 of these different applications that we

3:58 spoke about

3:59 to really unlock its potential the first

4:01 of these is network slicing

4:03 traditionally we've offered networks

4:05 which are one size bit all

4:07 and they've not really been dedicated

4:09 for service differentiation inside the

4:12 you know for different kind of use cases

4:14 with network slicing we have the

4:16 opportunity

4:17 to build dedicated end-to-end virtual

4:19 networks with prescribed cl

4:21 slas for all the different services so

4:23 whether you have healthcare or

4:25 industrial automation or

4:26 a v2x application you can allocate

4:29 dedicated slices with prescribed slas

4:31 for those

4:32 which will offer the real

4:34 differentiation that is required for

4:36 these services

4:38 if you look at some of the studies that

4:40 have happened operators are looking at

4:42 network slicing as the one of the key

4:44 ways to monetize 5g

4:46 and this report from ericsson is kind of

4:49 a testament to that fact

4:50 what ericsson is predicting is that 34

4:54 of the revenue growth that these service

4:56 providers are going to have

4:57 till 2026 is going to come from

5:00 non-best effort slices so only once

5:02 service operators start to launch

5:04 network slices which are dedicated for

5:06 services that's when they will see this

5:08 uptake in revenue and 34 is a huge

5:11 potential for growth

5:12 for service providers the other key

5:14 technology that's going to be

5:15 a differentiator in 5g is open ramp

5:19 so we all know that one of the key

5:21 motivations for open brand is to reduce

5:24 the vendor lock-in that exists today

5:26 inside the land domain but one of the

5:28 other key things

5:29 that open ran the opportunity to launch

5:31 new services and monetization options

5:34 the adoption for oren is expected to

5:37 grow almost

5:38 10 fold in the next six years we've

5:40 already started to see some

5:41 initial deployments happen in operators

5:44 but currently a lot of the operators are

5:46 in trial phases to look at how open run

5:48 applies to their technologies

5:50 if you look at the use cases they

5:52 include traffic steering

5:53 brand slice optimization massive minor

5:56 rent sharing energy efficiency and a lot

5:58 of these different use cases

6:00 now another key component within open

6:03 ram

6:04 is that how do you monetize services

6:07 using openran openland is great because

6:09 it helps you

6:09 unlock innovation it helps you unlock

6:11 you know the vendor locking so that

6:12 newer players can come in

6:14 but how does it enable monetization so

6:17 for monetization to happen with openran

6:19 we need a rick a rand intelligent

6:22 controller

6:23 and this brand intelligent controller

6:25 basically provides the operators the

6:27 levers that they need

6:28 to be able to fine-tune the rand network

6:30 based on the requirements that they have

6:32 for the specific services

6:34 or the network types etc if you look at

6:37 some of the recent trials that have

6:38 happened vodafone for example

6:40 just concluded a trial where they used

6:42 the random housing controller to

6:44 increase their

6:45 sell side capacity by two times in one

6:48 of their massive minor

6:49 use cases similarly kddi last year

6:52 demonstrated how the rig

6:54 is going to become the control point for

6:56 them for network slicing within the rank

6:58 domain

7:00 so with that said let's now look at what

7:04 is the impact that these new

7:05 technologies are happening are having

7:07 on the service provider networks on the

7:09 operator networks and how we expect this

7:11 to evolve

7:14 so definitely with with all of these

7:16 changes that are happening there is a

7:17 shift in the architecture that's

7:19 happening in each of the domains inside

7:21 the operators

7:22 so including the ram the transport as

7:24 well as the core

7:25 if you look at the ram we spoke about

7:26 the open brand that is you know enabling

7:28 new interfaces to be formed

7:30 the cell side is becoming significantly

7:32 simpler because we're reducing some of

7:34 the active components on the cell side

7:35 and the components like the control unit

7:38 and the data unit are getting

7:39 distributed much deeper inside the

7:41 network

7:41 and this also has an impact on how the

7:44 networks get designed

7:45 uh front hall and for mid-hall in terms

7:48 of the requirements for the network in

7:49 terms of timing latency etc

7:52 and last but not the least as we said

7:54 for the rank a key portion

7:56 is how the architecture is going to

7:57 change to adopt the ram intelligent

7:59 controller so that you can have

8:01 enhanced intelligence inside the rank

8:04 from a transport perspective the focus

8:06 is all around reducing the protocol

8:08 layers to make the transport simpler

8:11 and to virtualize and disaggregate

8:13 wherever it makes sense

8:15 the last but not the least from a

8:17 transport perspective the key trend that

8:19 is that we're going to see is a shift

8:21 of the traffic and service delivery to

8:23 from

8:24 the edge or traditionally from the board

8:25 towards the metro and to the cloud

8:27 so the metro is going to become a

8:29 melting pot of where

8:30 all of this is going to come together in

8:32 terms of the integration to the cloud

8:34 the network slicing and all of those

8:35 different pieces

8:37 and last but not the least is the core

8:39 where the architecture is definitely

8:41 changing towards a more microservice

8:43 based

8:43 service based architecture and the

8:45 distribution of the core towards

8:47 into multiple points inside the network

8:49 using the distributed upi for the mac

8:53 so with that said in the rest of the

8:54 section we are going to focus

8:56 more around the rand and the transport

8:58 pieces and with that let's dive in a

9:00 little bit deeper

9:01 into what does it take to make all of

9:04 this happen end-to-end

9:07 so if you look at it this from a network

9:09 slicing orchestration perspective

9:10 because that we see

9:11 is going to be the key driver for the

9:14 architecture change as well as for how

9:16 services get delivered so let us look at

9:18 this from the perspective of network

9:20 slice orchestration

9:21 so at the highest level we see that just

9:23 below the oss and vss we have the always

9:25 orchestration layer

9:26 and that service orchestrator is then

9:28 talking to the specific domain

9:30 controllers in each of the domains

9:32 so the rand domain controller

9:33 orchestrator for the ram

9:35 the transport domain orchestrator for

9:36 the transport and similarly the

9:38 the core domain orchestrator for the

9:40 core now let's

9:41 peel the onion a little bit more to look

9:43 at how this looks like from a network

9:45 slicing perspective

9:48 so we can see here that the service

9:51 orchestrator

9:52 has a slice management function and then

9:54 that slice management function

9:56 is also translated into each of the

9:58 domains so each of the respective

10:00 domains have their slice subnet

10:02 management functions

10:03 plus they have something that's a

10:05 controller for that specific domain so

10:07 for example in the in the rand domain

10:10 you can see that's the non-real-time

10:11 rate

10:12 in the transport domain that's the van

10:13 sdm controller and then in the core

10:15 domain we will have the nfv over there

10:17 so

10:18 each of these domains are going to be

10:20 responsible for

10:21 the orchestration and for the slice

10:23 management within their own domains

10:25 and they will be talking to the service

10:27 orchestrator that is responsible for

10:28 stitching all of this

10:29 end to end so let's now delve a little

10:33 more deeper into the rand domain

10:34 and then from there on we'll move on

10:36 towards the transport

10:39 so as you can see the rank domain if we

10:42 blow this

10:42 out we see that it has a few key

10:45 components the first is the service

10:46 management

10:47 and orchestration which is the smo and

10:49 the smo has a few components like the

10:51 management and orchestration

10:52 the oam the ram uh slice management as

10:56 well as the non-real-time grid

10:58 and then we have the oh cloud which is

11:00 the layer that

11:01 provides the containerization and

11:03 virtualization for all the applications

11:05 that are going to sit on top

11:07 so let's now look at how all of this

11:09 comes into play and

11:10 and how this whole slicing orchestration

11:12 works from a random perspective

11:15 so the first thing is from the

11:16 management and orchestration perspective

11:18 we talk to the old cloud

11:19 and the o cloud brings up the respective

11:22 control plane or the user plane

11:23 functions like the

11:24 po control plane as well as the near

11:26 real-time brick

11:28 now as you can see the rig has two

11:29 components the non-real-time rig that is

11:31 under the smo

11:33 and the new real-time brick that is

11:34 sitting next to the cu control plane

11:36 and each of them have their own set of

11:38 apps so we will discuss this in a little

11:40 more detail in the next slide

11:42 but the key thing is that the management

11:43 and orchestration informs the o cloud to

11:45 bring up the control plane as well as

11:47 the

11:47 near real-time break in the next phase

11:50 the operation operations and maintenance

11:52 talks to these components to make sure

11:54 they have been

11:54 they've been brought up appropriately

11:56 based on the requirements

11:58 and then collectively with the rand

12:00 slice management

12:01 the the user plane is set up based on

12:05 the different network devices so this

12:07 includes the

12:09 cu user plane as well as the the du

12:12 functions

12:12 now as you can see these functions can

12:14 be distributed into multiple

12:16 clouds regional cloud edge cloud cell

12:18 site and this would depend upon the type

12:20 of deployment

12:20 that operators have but overall the

12:23 model is flexible

12:24 enough to support these kind of

12:25 deployments now once we've done this

12:28 we need to tie this back to the

12:30 transport domain

12:31 and how does these slices translate into

12:34 what's happening in the transport domain

12:36 so for that purpose the ranked domain

12:38 orchestrator or this smo

12:40 talks to the transport domain

12:41 orchestrator to orchestrate the

12:43 respective

12:44 tunnels that are required in the

12:45 transport to make all of this

12:47 these slices happen and aditya will talk

12:49 more about how

12:51 each of these tunnels come into play how

12:53 are they formed and then how do

12:55 they talk they help these clouds to talk

12:57 to each other with the clouds between

12:59 the cell side and the edge cloud the

13:00 edge cloud and the regional cloud

13:03 so let's look at it a bit deeper into

13:06 the

13:06 this non-real-time break and the near

13:08 real-time break

13:11 the non-real-time break as we mentioned

13:13 it resides inside the smo

13:15 and it has yeah it hosts the r apps so

13:17 the job of the non-real-time break

13:19 is to provide policy-based guidance it

13:22 tells the near real-time

13:24 rick what it needs to do and the near

13:27 real-time break

13:28 sits very close to the edge cloud it

13:30 resides inside the edge cloud very close

13:32 to

13:32 the cu control planes and it hosts the x

13:35 apps

13:35 and this is where the what gets

13:37 translated into the how

13:39 so the near real-time rig talks to the

13:41 ceos and the dus and translates the

13:42 policy guidance that it receives

13:44 from the non-real-time break into the

13:47 actual commands to the

13:48 cu and to do you to make the actual

13:50 functions happen

13:52 so in the end when we want to realize

13:54 the use cases for the

13:56 rand which includes the slice

13:58 orchestration or

13:59 the massive mimo or the other use cases

14:02 uh like rank sharing

14:03 then this is the non-real-time break in

14:05 the near real time will have a

14:07 very important role to play to make this

14:10 happen

14:10 and the key thing here is that both of

14:12 these support the hosting of

14:14 third-party apps the rx and the x apps

14:17 which is where the most most of the

14:18 innovation is expected to happen

14:20 so from a juniper perspective juniper

14:22 has acquired a company called nexia

14:24 and we will be building both this

14:26 non-real-time as well as the near time

14:29 non-real time in the near real-time rate

14:31 as well as the smo functions

14:33 so that is how juniper is contributing

14:35 from an orion perspective

14:37 in this domain so now we've discussed

14:41 how things look like from a rand

14:43 perspective aditya

14:45 so what how do you think this translates

14:47 in terms of requirements

14:49 for the transport domain so is the

14:52 transport domain

14:53 the way we have it today is that good

14:55 enough or does that require

14:57 some kind of a change yeah thanks

15:00 so let's start with the base ib

15:02 transport infrastructure

15:04 there is a significant shift required in

15:06 terms of the transport

15:08 architecture to address the 5g

15:10 requirements

15:11 and so what are those requirements first

15:14 5g services and cell densification now

15:17 this brings a massive

15:19 bandwidth and scale increase majorly in

15:21 the metro

15:22 or xor networks this also brings

15:26 a massive explosion of end user id

15:29 devices

15:30 whether we have these devices as

15:32 enterprise cps

15:33 iot devices or consumers having said

15:36 that

15:36 timing has become much more stricter

15:39 with 5g

15:40 and low latency is now a key requirement

15:43 another requirement which we see for

15:46 this architecture shift

15:47 is around service differentiation and

15:50 application way routing now this is a

15:52 basic requirement when we talk about

15:54 network slicing

15:56 meanwhile we also see a need for

15:58 centralized

15:59 automation control and distributed

16:01 performance

16:02 which requires a hybrid architecture

16:05 how do we address these requirements we

16:07 need an architecture which is scalable

16:10 stateless and most important it should

16:12 be simple to operate

16:14 having said that segment routing if you

16:16 see segment routing is a foundation

16:18 for the ip transport architecture it's a

16:21 key enabler

16:22 that can reduce our protocol stack by up

16:25 to 70

16:26 which reduces the state and the scale

16:29 requirements

16:31 it is also inherently programmable which

16:34 kind of provides a centralized

16:36 automation control

16:38 now within segment routing there are two

16:40 approaches with similar benefits

16:42 to achieve the end objectives of 5g sr

16:46 mpls which is segment routing for mpls

16:49 now this is best

16:50 suited for existing mpls networks

16:53 as it brings the least disruption and

16:55 also mitigates risk

16:57 during migration to sr or the evolution

17:00 to sr

17:01 now within sr and pls we also have tools

17:04 to integrate with nodes and networks

17:08 which may not be capable enough to

17:09 deliver sr day 0

17:11 or which would require a certain

17:14 software upgrade

17:15 to deliver sr now the other flavor

17:19 of segment routing which we call as srv6

17:22 which essentially is

17:23 segment routing for ipv6 this is best

17:26 suited

17:27 for green field networks as it brings

17:30 the massive

17:31 scale of ipv6 and also

17:34 interesting capability to embed

17:38 the intent within the packet now from

17:41 services evolution perspective we see a

17:44 shift

17:45 more towards evpn which provides a

17:48 unified

17:49 services architecture for both layer

17:52 free

17:53 and later vpns while being transport

17:55 agnostic

17:56 which means it can run over plane ip any

17:59 flavor of mpls

18:01 or srv6 so this is

18:04 the basic ip transport infrastructure

18:07 requirements which we see

18:09 happening with the advent of 5g next

18:12 slide

18:13 nithin all right

18:16 so what are the key end-to-end network

18:18 slicing requirements

18:19 so within the ip transport world we have

18:22 been traditionally

18:23 delivering network slices as in layer 3

18:26 vpns or layer 2 vpns

18:28 however with 5g we see new requirements

18:31 which require networks to error to

18:32 stringent slas and

18:34 also very important a need to map

18:37 similar intent

18:38 within individual domains like ram metro

18:41 core

18:42 so these are essentially uh holistic

18:45 requirements

18:46 but what about the key network slicing

18:48 requirements first and foremost

18:50 traffic isolation now we need to

18:52 maintain traffic separation for

18:54 different slices

18:55 or customers and also a mechanism to

18:58 provide

18:58 security and segmentation another

19:01 requirement key requirement is topology

19:03 aware slice

19:04 now what do we mean by topology where

19:07 slice it essentially needs a slice

19:09 for example a load latency slice which

19:12 requires

19:12 a latency part between the consumer and

19:16 the application

19:17 and not necessarily the shortest path to

19:20 reach a destination as defined by igp

19:22 which essentially might not be the

19:24 lowest latency path

19:26 and one of the most important aspects or

19:28 requirement from network slicing

19:30 perspective

19:31 is the end-to-end orchestration

19:32 mechanism now this is the key

19:35 to maintain the intent which resonates

19:37 with consistent intent

19:39 within each domain this also unifies

19:42 the mechanism to deliver or provision

19:45 the intent

19:46 across different domains like the ram

19:48 the metro

19:49 the core and finally assurance

19:52 and guaranteed availability of network

19:56 resources

19:57 now this is a key aspect which is

19:59 requested by the slides

20:01 so these are essentially the key

20:04 end-to-end network slicing requirements

20:06 this is a public demonstration of

20:08 juniper

20:09 slicing solution and there is a url url

20:12 there

20:13 this provides you a view of how to

20:15 create and maintain

20:17 transport network slices using juniper's

20:21 nostril controller which essentially is

20:24 one of the most important aspects when

20:26 we talk about

20:27 automation and centralized control

20:30 you could visit the website for the demo

20:32 and the technical details

20:34 so now we have covered from a transport

20:37 aspect of the 5g and also the

20:39 requirements in terms of the network

20:41 slicing perspective

20:42 so nathan what's your view on the

20:44 end-to-end

20:45 slicing mechanism and how do we stitch

20:48 these

20:48 domains essentially yeah thanks aditya

20:51 so as we discussed we've looked at how

20:54 the ram

20:55 and the rig inside the ram is going to

20:56 take the responsibility of doing the

20:58 slices inside the network

21:00 inside the rand domain we've looked at

21:01 how the transport is going to look at

21:03 taking those that intent and translating

21:05 that into controller

21:07 requirements from a transport

21:08 perspective the key here

21:10 is that how do we stage all of these

21:12 together firstly from a multi-domain

21:14 perspective so that the intent

21:16 is from the rank or the transport of the

21:17 core we can ensure that that intent

21:19 is delivered seamlessly across the

21:21 entire network

21:23 and so that is the first component which

21:24 is where the service orchestration piece

21:26 comes in

21:26 but the other thing is that we also in

21:29 addition to being multi-tenant we also

21:31 want this to be

21:32 multi-domain we also want it to be

21:33 multi-tenant and multi-cloud so what

21:35 that means is

21:37 that in terms of offering these services

21:39 this could also be offered as a

21:41 tenant-based service wherein an

21:43 enterprise or an mvno might buy

21:45 a slice or a hierarchy of slices within

21:48 an operator network and they may want to

21:50 run it like that so you want

21:51 to be able to create network slicing not

21:53 just for your own operator requirements

21:55 but to be able to create it for hosting

21:58 other operators or other

21:59 enterprise services on top of that as

22:01 well the other thing is that we also

22:04 need to be

22:05 able to extend this not just into

22:06 distributed clouds within the operator

22:08 network

22:09 but also into the public cloud because

22:11 we expect that a lot of these services

22:13 that are not

22:13 latency sensitive will eventually move

22:16 to the public cloud and be hosted out of

22:17 there

22:18 so we need to be able to ensure that we

22:19 can do all of this stuff and do it

22:21 at scale so that is kind of like the

22:23 high level message

22:25 so with that let's wrap this up with the

22:27 three key messages that we discussed for

22:29 today

22:29 on how the architecture for for the

22:32 operator networks is changing because of

22:34 5g

22:35 and there are three key architecture

22:36 shifts that we're seeing the first one

22:38 is open rank which is uh making the

22:42 network simpler from a sell side

22:43 perspective uh it is making the rank

22:45 more intelligent by using things like

22:47 the brand intelligent controller

22:49 the second piece is the transport where

22:51 we are significantly simplifying

22:53 transport by reducing six

22:54 up to 70 percent of the overheads we are

22:56 making it more application aware and we

22:58 are

22:59 tightly integrating this with the ram

23:02 and the core domains to make this

23:03 an end-to-end solution and last but not

23:06 the least is network slicing

23:08 which has to be multi-domain

23:09 multi-tenant and has to be executed at

23:12 scale

23:13 with that we thank you for your time and

23:16 we hope

23:17 you found this session useful and we'll

23:19 probably see you in the next one

23:20 thank you thank you very much