Karim Rabie, ICT Chief Architect

Understanding Telco Edge

Telco Cloud
Karim Rabie Headshot
Tile slide with the words, “TELCO EDGE,” and a photo of Karim Rabie, ICT Chief Architect in the upper-right corner. 

The keys to understanding Telco Edge

What is Telco Edge? Watch this in-depth webinar with Kairm Rabie to find out. With practical content, key facts, and a case study, Karim breaks it all down for you. 

Show more

You’ll learn

  • All about the technology evolution justifying the rise of interest in Telco 

  • Guidance on how to interpret the Edge Definitions

  • An overview of potential use cases, Edge Cloud Stack, and deployment models

Who is this for?

Network Professionals Business Leaders


Karim Rabie Headshot
Karim Rabie
ICT Consultant 

0:00 this uh this month uh i'm going to talk about the the telco edge

0:05 um you have you have first before i start you have the quest the q a uh i would

0:11 say uh module you can just post your questions uh i'll try to stop uh like every like five

0:18 ten minutes to get the questions and and uh and have them answered if not then i

0:23 will collect the questions and then i'll post that as a as a post after the webinar

0:30 um just just quickly um for those who first time to say to meet

0:36 me i'm my name is karim raviyan i'm currently a chief architect from for one of the uh i would say the uh

0:45 giant telecom vendors my uh my my background is uh is actually mobile networks and

0:51 the uh and the transport so i had like previous roles in nokia huawei htc orange and stc

0:59 uh again sdc um i i see like i see the technologies from

1:04 two perspectives like um i'm uh i'm i follow the classical approach so whatever content i i say or i put in

1:11 a slide most probably it will have a reference and i will always refer to that reference so i'm a good i would say follower for the

1:18 for the 3gpp for iatf or itu and also i have some some good contributions uh in the uh standard

1:25 bodyside some and communities in addition to practicing the technology uh you know working many years in

1:31 defender weather side and also working on the operator side so also i see the technology

1:36 from you know the customer and the vendor side so um i i try to make slides that actually uh

1:44 i would say uh have have again have a reference from from whatever standards or

1:49 community groups but again i'm trying to put some practical uh i would say uh content and some uh listen word and

1:58 key facts um let's start just quickly again we have we have one hour so um

2:05 i i prefer some sort of a storyboard okay um and they named that road to edge okay

2:11 i'm trying here to explain how did we as a as a as a telco community or

2:17 or a service provider or csps or vendors how do you reach like a conclusion that we need to go to

2:23 edge and actually what is the edge so uh as a start i would like just to explain the the legacy

2:28 mobile network for those who uh i'd say not maybe familiar with that because

2:34 that's that's our starting point so um normally and here i picked an example of

2:39 of a typical uh you know for gsvg uh mobile core so we always divide like the like the

2:46 the the network into two port two imports the axis and the core and that's actually the legacy names of

2:53 the teams in any in any csp so the axis is always like the radio part the e note b the node b

2:58 the r and c these are like uh the the modulus uh constructing the radio and the accent access port and

3:06 you know it it may take other names right in other department in the other csps like it can have the access

3:11 the ran sometimes the bss so that's actually the name of the teams governing these modules

3:16 and we have the core uh delivering the core services as you see here in this light blue gather

3:21 like mme edges and escape gateway and you always have some sort of of uh connectivity requirements between

3:28 these modules either between the axis and the core between the core itself between the axis by

3:33 itself so these are this is what we call the transport or the network or the transmission teams

3:39 and it all depends on on what kind of connectivity that you are doing in which layer

3:44 and for example you can have a csp doing rnc sjn integration via mpls

3:51 network or ssn mme by mpls so that would be the mps port or the network port

3:57 you can think of node b or rnc at the transmission part but for the at the end this kind of

4:02 connectivity construct different departments and construct functional scopes

4:08 other than that other than the core and access and transport and these are the backbone of any mobile operator there

4:14 are other domains and other important entities like the vast value editors

4:22 like the rss like the bss for service provisioning bss for billing for marketplaces for

4:27 each channel for sales so these are the domains that construct a legacy mobile network

4:34 so with that approach and specifically from the 4g era

4:41 and that's like typically 2009 we had like some classical segmentation

4:48 of the network we have the access and we have the core and we have the service either intra

4:54 services provided by csb or like or internet or youtube or whatever

5:01 what was significant in in that network that the code network there was some sort of

5:08 split between control plane and user plane you saw this these nodes like mme right handling only

5:14 signaling [Music] so it was some sort of split right but

5:20 there was no i would say split intra function for example

5:26 uh you couldn't have uh sgsn or as gateway c and sk2 u so at the end cp

5:33 and user print processing was all in in in in the network in one nodes

5:41 and the latency by that time and the requirements coming from the itu and even the practical latency uh

5:46 fulfilled by the technology it had an average of 20 50 millisecond i made a little bit broad

5:52 because in many contexts there are different latency figures and even me personally from my experience i saw different latency

5:58 figures so i made like for 20 50 millisecond and and mapped to to what are there in the

6:03 itu and not what whatever in reality it makes sense so with that latency with that latency uh

6:12 there was no reason to go out the legacy sites or the legacy cos that had

6:18 been there central office so if you go back to southern nine we had like the concept of site right

6:24 or a concept of co where they told the core network like the sgsn usgs and the msc the media

6:31 gateway most of these were centralized and as you see because of the of the broad

6:36 latency requirements there was no uh real uh motivation to go out of the of the border of the

6:43 central office or or the classical sites on the national level or even in the regional level so that

6:51 was the case in in the 4g and by and by that time if you go to 2009 2010

6:57 the term edge was not there we were always using the axis and again the

7:02 access was pure radio and the core network kept their place in the classical sites

7:08 in the national international travel level or central level

7:13 with some you know after a few years we had many steps at an evolution uh

7:21 one of the biggest evolutions that we we saw uh in the in the mobile broadband or the mobile core with the

7:27 introduction of the cups or the cups control plane and user print separation

7:32 in 2017. so in 2017 and that was uh in r14 which is actually

7:41 the release that was a pre-5g i mean that the it's one of the four g3 releases

7:46 that was before the 5g the concept of cups was it was introduced this mean that if you

7:51 see the third bullet that you can have the node splitted into sk2 c as between u p gateway c p

7:58 3u tdfc which is a dpi to the fu so we had this kind of split and it was

8:06 a sign that that once you do the split you can easily deploy or position the splitted part

8:13 in any dc right so there is no coupling or obligation that you split and you put the user

8:19 plane in the same site of the control plane you can easily position the user plane at any site

8:25 keeping that the interface between them or the protocol between the control pane and user plane was standardized by 3gpp by the pfcp

8:33 protocol so you see it was a hint okay that you can do the split

8:40 and by that time in 2017 that was very uh really coupled with the concept of

8:45 edge y because back in 2015 there was an evolution of the

8:51 standardization of the hc mech okay european telecommunications standardization institute

8:56 they came with the mech isg in 2015 but by that time because the operator

9:03 didn't see any kind of compelling reason to to adopt the the new uh the new industry and go to the

9:10 edge right because there was no use case as i told you the the the set of services of 4g

9:16 and the end-to-end latency was very sufficient to be handled by the classical architecture so in 2015

9:22 the hmac was there also in 2015 there was an initiative called called

9:27 code code was a central office restructured

9:32 or re-architectured as data center so from the name it was an initiative to move the

9:37 classical co into a data center so that was in 2015 and again in 2015 most of us didn't see

9:46 any compelling reason to go to the edge because there was no business use case and the investment was huge but when

9:53 2017 came and from technological perspective we were able to do the split

9:59 this encouraged this encouraged the the the telecom tribe to start uh

10:06 considering the next wave what is the next wave it was actually the 5g so if the 4g didn't have didn't

10:14 construct any competing reasons to go to the edge with the 5g and i'm bringing here

10:19 the famous uh uh int 2020 the 5g requirements from the itu

10:25 you so for example one of the aspects is the latency right so you all hear about the ultra latency requirements right

10:31 so when the 5g came and i put it a bit broad with one to five millisecond 5g latency

10:38 requirements for the ur llc ultra reliable and low latency use cases it was clear

10:46 that the classical architecture the classical architecture of the 4g will never serve this kind of use cases

10:53 right because you will never be able to get a millisecond or even 5 milliseconds to the to the national sites or the core

10:59 size and what even gave a boost for the direction to the edge that the 5g architecture

11:05 it came out of the box already decoupling the cp from the up the control processing from

11:12 user processing if now you review the 5g architecture you will find that there are nodes only doing control complete

11:19 processing and other nodes like upf only doing or handling the the the user the user

11:26 play so that was really uh one of many motivations to go to the

11:33 edge and that encouraged the the operators to start uh reconsider or embrace

11:41 the mech idea that came in 2015 and by that time there was no adoption so i'm just

11:47 getting in the definition from the hc mech so in a nutshell

11:52 mech or the multi-edge computing from hd perspective making use of the split of the core

11:59 network so it's pushing the gateway or the user plane on the edge together with a set of edge

12:04 applications with an intention to have a local breakout local record it means that the user

12:10 the user will start to get the service at the edge level and for that time

12:17 for that time the impression for that time that edge is the access well in the next

12:24 slide i'm going to show you that it's not always the case but that was somehow a quick

12:31 i would say preview of it why there was um how did we reach a conclusion

12:38 that we need to go to the edge why was not this was not there in the 4g and why it

12:44 is somehow coupled with the 5g right you always hear the edge with the 5g but there were many motivations that were

12:50 there good efforts from hcmac uh however i will not comment about if it's adopted

12:56 or not but there was a good uh contribution from its mech to push the telecom tribe to go to the edge

13:02 the court also uh give give a good perception about how to to change the classical cos into data

13:08 centers and then with the 5g cloud native architecture and having this kind of split

13:14 and you all know that based on containers so it was easy to group all these facts and have a definition of the edge

13:20 that i'm going to explain in the next class so um let me stop here and have some questions

13:26 and come back to you

13:38 okay so there are some questions about the availability of the slides and the record so this time the slide

13:44 will be available and the record will be available as well

13:50 uh can you tell reference for one to five milliseconds latency for 5g mainly the itu the imt 2020 plus

13:57 [Music] plus some other some other content from gsma and from the vendors

14:03 but the the exact number is in the itu that's the reference for the latency for the 5g in the ipo imt2020

14:11 okay so let's go back now to the the

14:17 let me just uh close this questions and then we proceed so i answered this

14:24 one i answered this one then this one how applications deployed at the edge

14:30 adapt as per radio conditions change that's a very good question and it's a good coincidence that i'm answering that

14:36 in one of the slides okay so we continue

14:46 now very very important concept and and that's in my opinion

14:53 this is how you can deal with any any definition in the community there are normally two categories of terms of

15:00 technology terms one is that we can say that it's a standard term and one is a community

15:05 term okay so standard terms you know there are some specs in in 3gpp right and even in hcnfv

15:14 just doing definitions okay and that's why when you tell me nfvi for example okay i

15:20 can tell you nfvi it is the totality of software and hardware infrastructure that construct the environment to do host

15:26 vnf okay that's the definition in xc nfv so that's a standard term right i ran a code so these are all standard

15:33 terms and there are other terms i can i name them community terms

15:39 community terms this this these are again technological terms or technology terms

15:44 but they are coming from uh vendors from community from open source groups

15:49 from influencers from webinars right so it all the these terms they don't have a

15:56 standard definition but it all depends on the context that they were used in

16:01 and i can give you an example of a cloud native right you can see a lot of people saying cloud native right

16:07 but what does he mean by cloud native like for example is it is it like a stateless uh network

16:12 function like a vnf or is the mass to be on containers so what about vm stakes vms

16:19 is about the architecture internal architecture having load balancer so cloud native is a community term that

16:26 actually having a definition coming from from influencers from from the from the

16:32 organization so edge is exactly a community term and the next slide

16:37 i'm going to give you um the flavors of the edge okay all the terms used by by the

16:42 by the by the trap by the uh by the communities on the defined edge so you can hear a

16:48 lot about telco edge enterprise edge near edge far edge deep edge device edge so it's a community term so

16:55 we have to be flexibly welcoming all definitions we need just to understand the context

17:02 okay so i hit i just here give like an x edge

17:08 and you can just remove x and put emitter so there is a way or there is a mode

17:14 let me use the laser here there is a mode of naming edge with respect to the data

17:21 center positioning and i'm sure you all heard about the near edge

17:26 far edge deep edge and then device edge so it's like a hierarchical uh

17:32 positioning and then giving names to near far deep but if you ask me if you ask me far or

17:39 deep well there's no standard definition but you can see architecture and say i tried okay this is far and

17:46 that's nearly because that's nearer to the core side so but it all depends on the context that's why it is hundred percent a

17:51 community term and there is a kind of naming when uh people just start to to name the edge

17:59 with respect to the surf domain or the building in this case right so you hear a term like floating term like

18:06 terraco edge so when someone take telco edge you start to think ah okay most probably it is named telco edge

18:13 because it holds telco application right enterprise edge ah okay so that's the edge

18:18 providing the b2b or the enterprise port operator right so that's the edge belong to the operator not for example the

18:24 public cloud provider so that's that's the kind of naming that is there in the community and then there is a category that

18:30 includes all the community and marketing terms right like the term edge box it can be

18:35 anything in a box it can be an enterprise edge you have the naming from the from the for example hyperscaler and

18:41 from the vendors aws wavelengths so events is a solution of the edge for amazon

18:47 you have the itc mac so hdmi that's the the vision of hc to the edge computing

18:53 uh m is the multi-axis edge of computing and then you have another initiator from

18:58 gsma which is multi-operator edge having an edge common edge for multi-operators and there are a lot of

19:04 other terms and there's always you know the the hybrid mode right so you can see

19:10 in an announcement right that uh operator x succeeded to have a terkofar edge

19:16 enabled by aws wavelengths so you have a lot of definitions okay you have a lot of definition of terms

19:23 our duty is just to understand the context what does it mean because if someone say

19:29 far edge this doesn't mean anything this doesn't mean anything you should interpret the architecture and

19:36 understand ah okay why it's named far because maybe that's the the farthest point that they can

19:41 go okay and the near edge okay that's the news that's centered to the original ones so i made this

19:48 slide to give you the message that there are a lot of definitions and you should uh you should focus on the

19:55 context where this these definitions are being used

20:02 so typically typically and from from our experience working on on the

20:08 telco edge this is most of the cases belong to what i'm showing here but

20:13 again i admit that there can be other use cases not following the use convention so let me

20:19 start from the right to left what we have i mean without any edge in the story what we have

20:25 currently classically in any csp is some sort of core sites or central dc's or national

20:31 disease so you may you may call it site and when someone say site this gives some sort of a legacy

20:37 impression like that's a classical uh site that used to host you know tdfs

20:43 uh and uh media gateways and this kind of classical things working with sig tran and this kind of

20:48 ss7 and stuff but nowadays these sites are

20:54 are either replaced or restructured as a dc so you always hear the term central dcu national dc so

21:00 these are like the backbone of the network hosting most of the core functions and then going

21:08 a little bit to the border or to the edge you always hear the term near edge

21:13 so near edge in a simple definition that's the first hop after the regional and central dc's it

21:20 may map to the aggregation points the old t's it all depends on on case by case on the operator side but

21:26 normally when someone say near edge that's the nearest point to the original and central dc's

21:32 and at the end you have a use case for example if you are going to sell a certain product to the enterprise

21:39 segment and that then that use case required at a millisecond latency

21:47 well if the original dc's can fulfill the 10 millisecond voila you have the case served by the

21:53 region if you have the use case served by the near edge then use the near edge so that's why i always recommend before

22:01 giving any names before doing any investments you have to understand the anatomy of your network you have to do

22:07 an end-to-end latency assessment to understand what is where is the spot

22:12 or where is the part or what is the point of presence that you will deploy your wretch right because

22:18 you might not need to go to a deep edge while you're near edge you can fulfill the use cases right

22:24 okay so near edge maps normally to the aggregation and the rt the term forage

22:31 and sometimes people use deep edge in the territory context commonly refer to cell sites or the or

22:37 the or the pawns so that's that's a far or deep direction from the telco perspective

22:44 because normally normally the telco port

22:49 starts with one hop after the terminal so you know even people working on the

22:55 mobile network they know that so we always you know like draw this mobile and then whatever comes after the mobile

23:00 that's the telco port so starting from the access or from the cell site that's the deepest thing that telco can can go

23:07 so commonly the near edge on the far edge construct what we can call a telco edge or

23:13 provider edge that's a common case but i'm sure there are other contents

23:19 from other maybe vendors doing different tracks but that's from my experience and from my my uh perception to and

23:26 i would say assessment that i did on from various direction in the market

23:32 other than that we have the enterprise edge so the enterprise edge think of it it's like an edge as soon as in a box

23:39 for example you can you can leave a certain enterprise like a thin cpu sorry a synth server that include a

23:47 full private rg so that's that's a kind of of an enterprise edge the ucpe you know as part of the sd1 uh

23:54 use case the universal cpe kind of cpe that can that can host application that's a kind of interpretation and then

24:00 we can go you know you can go a little bit more deeper and you go to the device edgy itself and that typically

24:07 uh the iot gateway the devices and some some people put also ucpe as part of the device edge and there you

24:13 can start to hear terms like fog computing but i would say fog computing is not really

24:20 part of the of the teleco uh uh framework because again as i told you most

24:25 probably the teleco uh momentum stops either on the far deep edge or on the enterprise

24:32 interface edge so i end with one slide and then i'll go to the questions

24:37 i just brought a simple definition that i personally like okay and that was

24:44 i think mentioned in one of the blogs of the hebec uh uh i mean three years back instead of

24:50 18. so that was that was from alex rosnick the chairman of the mech rs3 in 2018.

24:56 simply simply edge is any location in your network where compute and related infrastructure

25:03 platform software are offered that is closer to user than your default center so you you

25:10 remember the default center which was original and central so if you go one hop near to the user

25:17 then you are in the edge and that makes sense right because near

25:22 edge far edge enterprise edge device edge you go one hop below your original central disease

25:28 that's the edge so it's a very simple definition so if you go closer to user other than

25:37 going beyond your regional center dc then you are you are in the edge so let me stop here

25:42 and have some

25:54 questions okay uh let me take one question does

26:01 the mic is standardized yes there is there is a full um there is a full isg industry

26:08 specification group for in in the mec uh in ic mech so hc is doing that

26:18 uh there's a question about will there be a one-to-one mapping between g node b and upf to achieve edge i think

26:26 i i will answer that in in this slide because the edge is a perfect fit uh for the ceo and eu

26:33 you know coming from the open run technology so so most we see that

26:38 most probably the end-to-end slice for example will host a c user plane and a upf so yeah there

26:44 is a c user plane and upf somehow uh mapping in some cases

26:53 and then a question about the cups

27:00 about the complexity of the cups okay and there are other questions so

27:05 please just because we need to you know i don't want to slip the i don't want to uh

27:12 because you know we have 30 minutes so let me ask about let me answer this one so the question about the cups about the

27:17 complexity of the cups in the lte and umps

27:23 so guys there was no real adoption in the market for the cups

27:28 right cups was there in 2017 but it was seen as a 4g feature and by

27:34 that time most of the operators said okay we have the cups but we don't have use case and um we are

27:41 waiting for the 5g so why can't we do like a full 5g cups like we because we all know that 5g is coming

27:47 with with with an implicit separation between control plane and user plane so if you ask me

27:54 people were not keen to to have the 4g cups and have it in the edge because if you go to the edge then it

27:59 makes sense to have the 5g order not the 4g so it's just a quick answer that

28:04 yes the cups was were complex and vendors took much time to support

28:10 why because their r and d were focusing on the 5g so if you are in 2017 and you have a 4g

28:17 feature called cups and in 2017 that was like one of the certain points for the 5g so most of the vendors were focusing

28:24 on the 5g and that you know motivated the operators to apply the caps or to

28:31 have the cuffs as part of the 5g and not not not by the no not not in the 4g

28:36 context okay so turquoise environment

28:42 and and by the way this slide is is is extracted from a gsma cloud reference model

28:49 that i i i participated in so what is the message here the match is

28:56 that okay the turquoise let me use a laser pointer

29:01 so we explained that the telco edge is commonly coupled with 5g it's not 100 coupled there is no really direct

29:07 relation but it makes sense right as you see from the storyboard

29:12 we had to go to the edge to fulfill the fuzzy use case because at the 4g time there was really no compelling

29:18 reason to go out of our classical sites and position anything out of it because we were 100

29:23 we were 100 percent delivering the 4g use cases so that's why it is seen as one of the ingredients of

29:29 the 5g because it has a good contribution and role in the ultra low latency and the emb

29:36 use cases if you go to the edge then you have to have a local breakout

29:41 termination and that makes sense to have that that justify why the ups the gateway you're there and why the

29:46 application is there because if you go to the edge the most probably you will have the low code breakout on the edge

29:51 so what i'm what i'm drawing here is just a collection of use cases and and

29:56 they're all explained in jsona for example telco edge is is 100

30:02 fit for the du and cu so if you have if you have a pop in your network a

30:07 position in your network that you can place the du and cu and then have the latency that

30:13 which we call the front home latency uh preserved then that's a perfect use case that's

30:18 that's a technical edge perfect use case but again it's not a use case that that will bring revenue right

30:24 when you when you onboard the the overran du and ceo in a in an infrastructure

30:29 that's fine but what are the revenue the revenue will come from the enterprise products or from the

30:36 slicing that services that you're going to sell on the edge for example you may start to sell for example

30:44 a slice for maybe a certain app so that slice

30:50 in a high level will consist of a cu user plane a upf and a third party application serving

30:57 the segment for example a slice for mining a slice for agriculture a slashing for for v2x okay so you can

31:05 play like uh um like like bricks and have the right offering for the

31:10 enterprise for the enterprise customers for example you can have also for consumers good uh good

31:16 use cases for analytics you can have a video analytics right on the edge you can also have a cdn

31:22 okay but regardless of the use cases what we see now in the market it's not actually about the application

31:29 itself it's about having a solution that may have some sort of a marketplace

31:34 that enable the user to go in a public cloud experience for example like the user going to amazon

31:40 and then select uh okay for my edge locations in arizona or atlanta i want this application then i click on

31:47 it and then i get on the edge serving my factory there so that's that's the new mentality

31:52 you you don't just sell sell a telco edge with with a upf in a box or five in a box no

31:57 you have to you have to give your customers some sort of vases and not static vases not static

32:03 value-added services but you must give them some sort of digital interface that that can allow them to go and deploy

32:10 deploy the application that fits for them and not only that not only that you see the pass

32:18 right in in the in the slide is a platform as a service pass is a

32:23 platform as a service think with me think with me

32:29 if an enterprise cast mood then he get he gets for example an edgy slice from a

32:34 csp and that edgy slice consists of of like a mini mini 5g at least user plane like

32:39 for example a upf and then a certain application from the csp to the enterprise customer

32:45 but if you are go if you are providing a profitable service to the interface customer the interface customer itself himself

32:52 can do can develop his own applications so if you are in on an aws for example platform

32:59 and you sell aws then the enterprise customer if they have the development skills of

33:05 aws and you see a lot of certificates you know on linkedin so people now are really

33:10 doing a lot of effort learning aws and being set as a developer so if the customer has this kind of of

33:16 skill set then the customer will be able to develop more application or more tuned

33:22 application to their needs so there are a lot of

33:28 use cases that can be enabled by the telco edge 100 it will host the network function

33:33 port which is the upf and the perfect fit for the oran do and cu

33:39 plus edge applications edge applications can be anything it can be anything there are many use

33:45 cases on the community internet like for analytics for cdn for v2x and for all industry signals

33:53 i will stop here before i you go to the edge cloud stack and have like a quick [Music]

34:00 review on the questions

34:13 okay so one one question one question about is actually the near-far enterprise edge

34:19 is all about latency or there are other use cases that have like other factors like security and ownership

34:24 and discounting and that's a very that's a very really good question so the first factor

34:32 for sure it is latency okay so latency is really a big factor on on determining the uh the pops location so

34:40 for example if if if it goes on like uh in a logical way then the business

34:47 department or the ebu and ncsp they start to think of digital products

34:52 enterprise products and then they go to technical teams to procurement tell guys we need this uh specifications or this kind of

34:59 capabilities to be able to deliver these these products okay and then the technical team they do the assessment and say

35:05 okay to deliver this uh applications or these products we need to have this kind of location in

35:10 that that edge near or far edge but there's there are other factors okay determine

35:16 where to position the edge for example in the middle east in the middle east there are still many

35:22 many limitations about having uh an on-prem deployment at the

35:27 cloud premises for example you cannot give a private 5g to the enterprise customers right

35:32 uh some other regulations for example that you cannot uh host two data for two private 5gs at

35:41 one physical host okay for to avoid the uh to have this kind of security isolation

35:47 so that's why the telco edge is the most favorable uh uh choice from security perspectives

35:56 because you still own you still own your data it's still on your premise you haven't

36:01 gone to the level that you go to an enterprise edge or to even even far beyond that so

36:08 yes latency is one of the main factors but again security ownership the operational model is also

36:14 other is also one of one of the other factors so i'm going to talk about the edge card stack from top from bottom to

36:20 top so guys we all have you know cloud

36:25 deployments like on-prem teleco clouds iot clouds private cloud so what what would be the difference now when you go

36:31 to have this kind of deployments on the edge so from hardware specifications we

36:37 started to see uh many many aspects that actually differentiate the edge hardware specs from the others one of

36:43 them is the form factor right so we started to see like more models that have

36:48 uh for example a one u not not to you it's only one u we started to see like a very dramatic

36:55 decrease in the depth of the server so if i say average server can be 100 centimeter 120

37:02 right 80 centimeter so in the edge we see models from from from

37:08 different hardware manufacturers that can go to 60 centimeter and 40 centimeter so the form

37:14 factor really is is is adapting to to adapt to the to the edge uh use case uh

37:22 from uh and even even the outer body uh ruggedized it can be like having uh

37:28 like a solid body to protect the server from from you know the environmental uh

37:33 conditions that may happen may happen at set site so there are a lot of you know new families from from

37:39 hardware coming from the new vendors sorry from the the well-known vendors like bell hp and

37:45 and even the new vendors edge is very coupled with the accelerators

37:50 right you people working on on that and that domain they always couple use cases with accelerators and for

37:57 example when you tell when you tell me gaming uh gaming then it's a gpu right okay we need gpu and

38:04 then uh oran orion fpga fpga uh caching persistent memory so there

38:10 are a lot of accelerators that that built that is there and built in the servers to support the the edge use cases so

38:17 this is this this is one of the aspects from security press practice

38:23 uh and people working people working on the mobile mobile domain will relate you know that

38:29 eno b itself the enol b and the 4g to be registered as as as one of the managed enough b's

38:36 there should be some security handshakes to make sure that that this erb belongs to us go up to the network right

38:42 so the same case in the servers there are a lot of security measures to be done on the hardware something like tpm for example trusted

38:48 platform module it's like a microchip in the server that that does this kind of of

38:53 of authentication to make sure that that server belongs belongs to us and not not an intruder in the network and then

39:00 nothing and the the performance and power so there's some

39:08 some requirements to have high performance on the edge for some use cases like video analytics artificial intelligence

39:14 engines in the edge but on the same time the power consumption and the requirement

39:19 there should be less than the core sites because there are no real power uh i would say sources like

39:26 same in the in the in the central debt center so we see in the market a degradation in the power requirements on

39:31 the server of the edge so with all these factors if you now go and then browse the data sheet of any

39:37 server provided by any vendor for the edge you'll find that really uh there are like major differences from

39:45 the ones that being used on the on the central sides on top of that as a cloud offering

39:50 what what what does the edge should should provide so initially we had the understanding

39:58 that that from cloud offering perspective we should support infrastructure service like vm-based hypervisors containers and

40:04 hypervisors and then containers environment but i made that in gray because nowadays we see a lot of momentum to go

40:12 for bare metal containers so i i kind of uh scanned all the 5g

40:17 solutions from all vendors so most of them are container all the edge applications are lightweight it's a bare metal container so you

40:24 should provide container as a service and then you should have an sdn capability and here i'm not talking

40:29 about a certain commercial product i'm talking about the capability of having a software defined networking

40:34 either it is provided natively by by whatever container solution is provided or by any kind of a back end that can

40:41 fulfill the software defined network and pass this is something that we are looking on

40:46 we believe that the solution on the edge should provide a platform as a service

40:51 and under the name platform and the servers each vendor defined that differently some people see that

40:57 if if you provide a solution for fmpm configuration management backup and

41:02 restore that's a kind of pass that's fine some people provide a real pass for for people to develop

41:09 their application so that's again a pass uh some provide database as a service for 5g so that's a pass so

41:15 but the pass as a capability should be there and then on top of that the ideal

41:20 application for the edge 5gc is one of them definitely for core the iot the vran co and du and then the

41:28 the cbn uh the the cdn as one of the uh one of the use cases

41:34 for consumers but i would like to stress on one point again i would say that

41:42 the certified application or the marketplace is really one of the main requirements now in the market nowadays

41:49 so i don't get a static stack as an enterprise customer and then i get a 5g code from your side and an iq

41:54 application no i need to have the flexibility at any time to start adding applications

42:00 and to mobilize application like now i have like a factory in toronto and then i open someone

42:05 on on in in india and maybe in egypt so i need to deploy

42:11 a full list of applications like you have to give me a complete portfolio and then in a direct provisioning way i can

42:18 deploy uh application in the in different different education and i will show you that in the last slide and that's that's

42:24 the the the the the main aspects about the aws wavelengths

42:29 offering and i'm giving the example with verizon in the last slide so that's that's an important i'd say

42:36 important uh brick in the overall edgy solution so do you have the stacks

42:41 then you need to have a management uh management stack or a construction stack so you can start to

42:48 have controllers these controllers either centralized managing remote

42:53 compute nodes or workers or it is distributed so controllers like kubernetes like europa stack like

42:59 whatever you can have orchestration stacks so that orchestration site can be a new

43:05 orchestration for the edge introduced as part of the edge solution or you can extend the capabilities of

43:10 the orchestration stack in central sites to the to the edge i'm going to show that in a while

43:15 and then you have to have a full operational tool set for fault management uh performance

43:21 management configuration for security backup and restore and this kind of operational tool most probably they must be integrated

43:29 with your rss for the holistic i would say visibility on the on the network so that's the cloud stack

43:39 we have to have 12 minutes so just a quick informative just informative

43:44 architecture of the edge host which is the edge pop from hcmec perspective so

43:50 you have here like a set of applications that are going to be on the on the edge and most probably you will have the 5g

43:57 user playing on the edge and you have something called mobile edge platform this mobile edge platform

44:02 is providing services for the applications so it looks like a platform service

44:08 thing it's not a directly mapping but there is a platform that is doing services for the applications and then

44:13 there was one question about how these applications get no get to know the radio conditions and how

44:20 they can adapt so for example from each from hcme perspective they said okay d

44:25 ucu okay that's a that's a perfect fit for the edge host right so why can't we do a logic

44:34 a logic here that this mobile edge platform can get feed from the ducu about the

44:40 radio condition radio congestion and then can provide this information to the application and based on that the

44:45 application can have like different logics or different behaviors so that's the idea from from from from

44:52 hcmac that you have something called mobile edge platform doing platform servers have some integration with the

44:58 ceo and du as one of the use cases and these apis are defined by the way so you get feeds and inputs about the

45:05 congestion and radio state and then that kind of of feed is conveyed to the edge

45:11 applications to have you know different scenario or different different you know different directions

45:19 i have to move fast i'm sorry

45:24 so we have three slides uh one of the well-known i would say the

45:31 uh not debate or confusion about the edge is actually how to start okay for for teleco operator for csps

45:39 so most probably two years back uh two two architectures uh

45:46 logics were proposed from the vendors one is called standalone edge clusters and standalone

45:51 edge cluster this this means that okay you have a centralized cloud that is that have controllers

45:57 and compute or worker nodes open stack or kubernetes or whatever and then on the edge you deal with edge

46:04 with each edge pop as a separate entity that has that can have like a controller or computer like

46:09 a mini mini controller and many computers so that that this work called standalone edge

46:15 cluster and let me just remind you that the investment on the edge can be huge because

46:20 edge can be like servants right of sights hundreds of thousands of dc's so

46:28 it's it's a painful investment okay so you cannot just go and invest without making sure that there's

46:33 a business case and there is like a well-defined revenue stream the other option

46:38 and the name here is coming mainly from red hat dcn distributed compute compute nodes this

46:44 means that okay you can have your central cloud having controller and computed workers

46:50 and you can start deal with edge as remote workers like remote uh remote compute nodes remote

46:57 worker nodes in case of of of kubernetes so these are the two well known

47:03 approaches there are maybe other approaches like uh they call the device edge which is

47:08 a one server that hosts everything but i'm just focusing on these two because the csps are always evaluating these two

47:15 uh these two uh these two models and i'm here doing an assessment

47:20 we can do it together actually uh the standard edge clusters and the dcm

47:27 so from footprint perspective csps see that okay i don't have to invest or

47:34 to reinvest on on controllers on the edge and because that that's most probably uh costly so

47:42 from footprint perspective the option of dcn is really much better where you can just deal with

47:47 any remote pop as just adding a new a new compute node you know expansion

47:55 from high availability and service impact in case of failure so you see that the failure of

48:02 centralized site in the standalone edge cluster this doesn't mean that the

48:08 edge sites fail okay i know that centralized

48:13 that center fails that that's a huge thing but let's think about of disaster recovery or a disaster mode dc is is totally out

48:21 of out of uh out of service so if you lose the central sides

48:26 this is you don't lose the edge while even the dcn if central dc controllers

48:34 is is out there you lose the connectivity with the edge this doesn't mean that the

48:40 service stops edge disease can still you know handle

48:45 traffic right in in a way or another but the communication the dialogue any kind of

48:50 of new adjustment any kind of governance is is lost that's why we see the edge the edge

48:57 cluster standard launch cluster as more i mean having giving better high availability from operational

49:03 perspective they believe that in the dcn approach because you have a central

49:08 brain and you have remote workers you can just do you know the automatic configuration the dtp the infrastructure

49:14 the code the expansions in a in a very uh classic way the the same we do it on on central disease

49:22 while in in the in the standard edge clusters people believe that you may have to go

49:28 side by side dc by dc if you don't have the automation framework in place so that's

49:33 why that this end looks a little bit better from scalability adding a new pop in a

49:39 dcn just adding a new compute nodes it's like it's a very classical uh process but adding a new

49:47 pop in the standalone edge cluster require uh structuring the full dc controllers and

49:53 worker nodes second thing is a brown field case and that's a very important most probably because most of the

50:00 operators they have on-prem clouds so they have a central uh stack

50:05 that have controllers right in the in the center sides so if they want to go to the edge as a

50:10 pilot or as a testing they always prefer the dcn because they say okay let's go

50:16 like having one or two three dc's okay capitalizing on the investment that we did on the on the

50:22 central side and let's see right so that's all the starting point of our brownfield uh cases while in in standard edge

50:30 cluster it's a perfect fit for for a green field but brownfield it's always preferred to go to the dcm

50:36 from operational tools of f caps again because the central dc's they have the connectivity

50:42 to the rss and bss so again most of the operators they want to capitalize on that integration because people working

50:48 on that part it's one of the lengthy painful integration to integrate with the fault management performance management

50:54 to the rss that's why from operation crude perspective dcn somehow

50:59 is preferred for interlocks with orchestration and to be fair and that's my opinion

51:05 it might look like dcn is a good approach because there are orchestration stacks are there and we can just extend

51:11 the orchestration capability on the edge however on the edge

51:16 it's it can be a sweet spot to welcome new players startups or new solutions

51:22 for orchestration that can integrate to the office at the end or the end-to-end orchestration or the multidimension or whatever

51:28 so that's why i didn't give a uh like a higher weight to someone i believe that both of them

51:34 are okay with terms of orchestration if it's a dcn then you capitalize on central uh

51:40 orchestration stack and you extend the capability of the edge if it's a standalone edge cluster then you have

51:45 like a new domain orchestration for edge maybe coming from a new player and then you integrate with the end-to-end uh

51:53 i have like four minutes i'll just finish by this one because it's very very interesting

52:00 so if you go to to study the case for example of aws doing a partnership with verizon

52:07 so first of all if you want to get this service you go to amazon so that's number one so you as an

52:14 enterprise customer you like to go to amd right because you are used to work with amazon so you can get hd services from amazon the

52:21 same way you you get your your cloud service the i mean from from aws so you go to aws and then you find that

52:28 okay aws is saying that we are there in that in these cities doing partnering with the csp verizon

52:35 okay so if you get the service from us then you get it from from verizon and from from the definition itself what

52:41 aws did that they went to the uh let me just find you the exact contest that they

52:48 went to uh the csp data centers at the edge so they went to communication csp that center at the

52:54 edge and they put their stack okay so that you can get a full solution

53:00 mixed by verizon offering an aws offering but what is the good thing and that that that uh what i was uh

53:08 uh stressing on that when you get aws wavelengths you see you get you get these right you

53:14 can easily you can easily get the machine learning uh seg maker right you can get the aws

53:22 robo maker so you get a full list of certified applications on the edge so it's not just a network port an

53:28 engine to part a 5g partner you get a 5g from verizon but you get the whole list from amazon

53:33 and that's amazon environment right we can do any anything under under the

53:39 the title platform as a service so that's that's the edge really that the hyperscaler they have that they

53:45 have a full stack of certified applications they have a global presence globally yeah imagine with me that if

53:53 this customer a service in atlanta and

53:58 he has like a branch in in south korea so he can get the same because uh aws is having the

54:05 same partnership with with an operator in south korea so he can get the same same edge service

54:11 in atlanta and in south korea so that's a global footprint and that's an edge of the hyper skills

54:17 on on that compared to the typical csp proposal right because for example if i am

54:23 giving uh i'm a vodafone uk giving a service most probably it would be in the uk right but aws

54:30 or hyperscaler helps csp to have this kind of global global footprint or global delivery

54:37 so guys i i i it's already one hour first of all i hope it was really uh uh

54:44 beneficial for all of us uh i tried to to to cover the topic from

54:52 i would say market perspective from sender's perspective and and actually giving you the hint on what's happening on reality and i end

54:59 on with the case study because that actually showed the trends about the edge so there are a lot of

55:06 questions strategical questions to be answered as a csp shall i expand my capability

55:12 from the court to the edge shall i partner with the hyperscaler

55:18 shall i go with the dcn approach shall i have like an on-prem deployment or i go directly for for enterprise edge

55:25 giving a private 5g on a box there are a lot of a lot of questions uh to be discussed and to be to be

55:31 answered so we have like i can take some

55:46 questions okay about question which of the vendors

55:53 are offering dcn so dcn mainly is well the term is coming from red hat so i would say red hat

56:01 the global footprint provided by hyperscatter is similar to gsma mom yes but i would say the different

56:08 yanni because the hype the the global footprint is actually there from from the hyperscale and gsma

56:16 is trying to having the same but but it's coming from the hyperscape

56:24 okay okay guys i would like i'd like lily really to

56:30 thank you i i i personally enjoyed this one it's it's uh it's always a pleasure to meet you and i i

56:37 really like the questions i'm sorry i couldn't go through all the questions but uh i promise you that i

56:42 will just collect all these questions and have it answered and that we can use it all as a reference as

56:50 a frequently asked questions guide for the edge

56:56 so if you have any any final comments i'm here uh yeah thank you

Show more