Red Hat OpenShift Telco | OpenShift Commons, 30 Nov 2018

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: OpenShift Commons Telco SIG Nov 30 2018 Full

Description

Telco on OpenShift SIG Meeting Nov 30 2018
Agenda:
Full Meeting Recording
Co-chairs: Diane Mueller and Paul Lancaster (Red Hat)

SIG Charter/Mission – Diane Mueller (Red Hat)
Introductions
What is new in Kubernetes Networking? Kural Ramakrishnan (Intel)
Containers at the Edge Azhar Sayeed (Red Hat)
Next Meeting: 4th Friday of each month work for everyone at 8:00 am Pacific - Jan 25th
https://commons.openshift.org/events.html#event|telco-sig-2|907

A

B

So I said: Corolla welcome my senior here yeah I.

A

C

A

C

I'm really looking forward to this curl, because I need a crash course on multiple things here, because that was the first yesterday that I heard of Malta, okay, good. It's very, very cool and.

B

Then, of course, we have our Red Hat lead on Montes thing is here as well. Welcome thing all right, hello,.

D

Good morning, I.

B

Guess we're looking for a Tsar to join and then we can go into the kickoff and things like this by.

E

B

Morning, I'm, just to let you know, Diane I did promote this widely internally on our NFE boards, and things like this I also asked folks to promote it with partners and customers, so I'm, hoping that we get quite a few folks. I, do see some very key Red Hatters in terms of both our CEOs office, as well as our telco architecture team attending. So that's great stuff.

C

So what in in the chat is a link to hack em video, which is what I use for meeting notes? If you can add your name to the list of any attendees I'll try and do it while we're chatting you don't have to I, should be able to get there well we'll get started in a few minutes. We'll give everybody. Let's say you were three more minutes just to join up, but I'm really pleased that Paul is the instigator for doing this.

C

I've been having a lot of side conversations with some of some of you and it's been one of the things I like to do. Make everybody a pop except me, so I'm just gonna. Well,.

B

That might be dangerous in my case, I think people who know me know that I can be a little verbose and so I'll try to keep it limited today.

C

One of the things so I'm just gonna get right into the sig charter admission, because it's really um there's a just I like to make clarify the distinction. I mean they occur, Bernardin working in technical sig and an open shift common sig and in the open shift common sig.

C

What we like to do is have sort of twofold: one is to have people share their use cases, their lessons, learn their best practices and that in this space and to get updates on new technologies- and you know roadmaps for new technologies that Red, Hatters and other Commons members are working on and using in this space Elco to be specific and the distinction between this and say a kubernetes community sig, which is really about.

C

You know how to add something to kubernetes and how to work or how to how to integrate something on new technology into kubernetes errs along that line. So we try and because there's some there's tons of kubernetes SIG's that these really are about sharing lessons learns best practices, your use cases and getting that fire hose of information about. What's coming down the pike and what's people are passionate about in new technology, so it's there's I, think I think that's a pretty clear distinction.

C

So, if you're looking to contribute code to you know some new project inside of kubernetes or around kubernetes or around openshift, we can find you those places too, but this is really kind of a conversational thing. If you have it so I just wanted to make that clear, and it's really about connecting peers to peers. So, though, we have a lot of Red Hatters on here, sometimes even at Red Hat with 12,000 employees, we need to connect to each other.

C

But mostly what we're trying to do is connect all of the the different members in the OpenShift ecosystem, whether their hosts service providers telcos end-users to each other, so that they can have conversations.

C

So that said, I'm just going to enter myself. I am the director of Community Development for Red Hat's cloud platform, which means all kinds of different things and I wear many hats. So please do ask me um if you are interested in operators, machine learning on OpenShift.

C

You know any any any of the interesting little things that on the side and I do a lot of the cross community collaboration with the upstream project, so I'm a good touch point for those things and I. Also what I'll probably do is everybody who's on here, I'll invite you all to join a slack channel for telco so that you can have telephone conversations, I run the slack or open ship Commons so that you know you can continue these conversations in the interim and with that I'm gonna.

C

Stop because I don't really like the sound of my own voice, heard it too much and let Paul talk and introduce himself.

B

Thanks Diane, so yes, so Paul Lancaster, with Red Hat I work inside of what's called the cloud platforms business unit.

B

So the cloud platforms, business unit encompasses several different pieces of technology, including open shifts or red hat, open shift, container platform, open, shift, online, etc and I've been working in the telecommunications ecosystem for a few years now you know working with I'd, say our most strategic partners and a lot of the folks that you that actually are on the call today so working on innovation for for our partners in order to deliver frankly, the best possible technology we can to the telco space.

C

Perfect so I I see if a number of you have put your names in as the attendees here Brad do you want to just say who you are and what you do at Booz, Allen sure.

F

Brad Appleman here I work in a research lab at Boot Allen Booz Allen were telco focused, my historically I'ma telecommunications engineer, somewhat new to the containerization environment, but we're currently running a a open EPC, a OpenStack platform or a partner with Red Hat and I'm interested in possibly we're looking at moving our open EPC solution into a containerized environment from the OpenStack, but for more of a youth I'm more of a use case engineer, not a software developer. So.

B

The open EPC, have you I mean, have you started the containerization of that? Would you be? Obviously, would you be looking at contributing that containerization of that opening PC back to the community so that anyone can use it containerized.

F

I, probably don't have that leeway with the Booz Allen place to do that. I.

B

Mean because one of the things that we're looking at is there's there's a there's: some open source software in the community like, for instance, clear waters, IMS software- that you know a lot of folks- are working on container rising right now and it's been containerized to some extent. But it's just always interesting to see what open source software people folks could potentially use- and you know, frankly, writing on openshift for calcio solutions.

F

Well, we're just starting out the gate on of that path and I intend to stay tuned to a red hat and openshift, and this the focus of this city gets a particular interest to me particularly interest to me, of course, based on what I information I just provided sure. So.

C

I see pasties name, so if you could quickly introduce yourself Pasi and then we'll go through that list, hello.

G

Yes, thanks Sara, my name is patty Brennan and I I mean the CEOs office in the Electra partner, engineering team and I work closely with author and others on the its use cases in general and radio access network and mobile network cases, especially. Thank you.

C

And thing yes,.

D

So from pan, I am from the city of office, so my team mainly works with cross team upstream multi development, and we are also responsible for delivering motors. They were in Novi shift.

C

A

I'm Sanjay igari I'm in the telco global architect, team I work with other and Ian on the top 25 accounts and.

C

H

C

Our you're, just because you're speaking to your microphone, is very low. Oh I'm.

H

Sorry, let me just try and fix that, but just.

F

Wanted to quickly introduce myself.

H

I'll fix the microphone later, as I said, I've been I mean I work in the same team as Sanjay I work very closely, with posi on the edge architectures and- and we will talk a little bit about that later on in this session.

C

All right, Alexander.

I

Hi I'm television, architects from Amir or covering few talk accounts and few colitis.

C

Alright and Emily.

C

E

Sorry yeah, yes, with the mute, I am Emily, yeah, I, work at red eyes and I've been working with sell cool. But after five years with the open ship and come here and you're.

C

A fellow Canadian correct, yes,.

E

C

I'm up in Vancouver are you in Montreal I'm.

E

In Montreal awesome so.

H

Just for everybody's, you know interest and really comes to us from well she's a recent hire into reddit, so welcome Emily. Thank you for joining me and.

C

J

William I'm a senior open shave solution, architect, the North America telco team in Retta and.

C

Then oh I see hide his drink is David ghertner I.

K

Am David I'm a consulting architect with the Indian telco practice that I work, basically with telcos across India in the implementation early most in OpenStack, but I get a lot of questions around. What's the what's happening with open shift and that's why I'm here right.

C

And then Andrew.

L

Andrew Toth same team as posi & Fung work with partners, so I have interfaced with Paul on a number of different partners trying to get their upstream development needs fulfilled in.

F

L

C

Then Hyde who's just joined us hi, hello,.

A

M

Hi I'm chief of the record Japan and the promoting the opens if the BTB takes platformer, that wedge together with telecom carrier, yep.

C

And I just put corrals name in there from Intel, so with Carrell like you're on yeah.

N

So this garage huh name is clément over krishna, so I'm working in cloud native orchestration team in the Intel. So they are basically from data center group within Intel and we basically work along with the customers and most of our use cases go to market. So we are working with Fang and team from CTO office, so we usually work closely and we are like close team together. So we will talk about this. One in my presentation here, perfect.

C

D

Good morning disappointed I'm a serious solution architect based in North America from Red.

M

C

I can see an aniket from juniper.

O

Hi product manager further the contrail product line at juniper networks, and we are very excited to see this telco signal. We have. We have a bunch of good developments to share awesome.

C

We'll put you on the agenda for the next topic, the next meeting, so that would be great, so I want to get started because we have two talks that I want to do and we were recording everything. So the whole video of the entire meeting will be posted on a YouTube channel and I'll, send it to the mailing list and put it in on the calendar for this meeting.

C

So you can find it so without any further ado, I'm sure, there's probably a few other people will join us as we go, so I will add them to the agenda Correll. How about? If you share your screen, I'll stop sharing mine on. We set it up and start your what is kubernetes and networking and why it matters to telco people.

N

Good go to guys able to see my screen. We.

C

Can see it and we're seeing the the outline version, not the full full version, so you need to go into presentation mode, yeah.

L

C

N

So hi guys I am working in BCG group within Intel, so we are working on in close partnership with the same team or from city offices. So I want to say this whole presentation is from the contribution from both the internet, the attack team and that's why I put input from both printing and Doc's mix and Billy. So, let's see what is new in kubernetes, my cooking, so before starting into the topics as traditional that everyone starts with what is cloud native I also going to start with? What is my big collaborative?

N

So my simple approach to what cloud native is cloudy with the approach of building and running application that fully exploits that going cloud. Computing, business, deliverables models and cloud narrative is how you create your application, deploy, maintain and it's not about where you place this application, so the application could be facing the edge code or in the data center clouds and cloud like belleville model is very appropriate for both public and the cloud private cloud segments and the key attributes.

N

So one of the key hat reduces like it should be stateless and they say length and it should be self feeling, and so this is a basic information. I want to begin with a cloud native and the next one. I want to little bit talk about the network or transformation from Intel's perspective, so the from Intel perspective, the you can see the broader network transformation for the modernized networks and within interview are organizing to either VMs or container. So we don't go into.

N

The argument is like, which is better VM nor container, but the fundamental is into low detection is bit for all the models and when we talk with the customers most of the customers want to have your mix third kind of so technology and deployment models at all the time. It's not all the time.

N

It's only containers in some cases it's kind of a hybrid model where they want to have both VM surface containers, basically it for the Bromfield deployments and not all the customers are willing to transfer to the cloud native so that that's why I want to iterate that we have a mixture customers who wants to work in both cloud native and also so the interesting point. What I want to point out here is like we are in the cloud ready, which means clarified.

N

We are not fully gone into the cloud native path going cloud native, so the path for the cloud. It is still that so we have to do lot of work to go towards the cloud native, so we are currently doing the cloud rain ready face actually.

N

So, let's jump into the actual technical stuff, so we will talk here about what is container networking and what is the various deployment models. So this is a slide which is kind of picked from the open, nfe architecture. So you can see there is various be enough applications. So we are working with closely with customers who have containerized to there be enough applications and we are working with various technologies like oviya, spiritual switchers, Fido, DPD, cave and also in common.

N

It is we have a CNA community and we also working with NB a porta textual like man, o and O Assam, so with Intel Xeon, like curves mixture of failures, authorization and service orchestration as well. So what does the things really concerned as when we talk with a customer? So one of the deployment models whenever the customer talk is both like bad metal models and unified infrastructure models.

N

So when you say look into the bare metal deployment models, the marrow or OS 'm is act like a service orchestration where the kubernetes, this kind of defect or auto station engine fits a talker at the runtime a container and time. And the thing is that the kubernetes as to manage both of compute network and storage, and then we talk with few customers who want to use the brownfield or bad. They have big burst to a lot on the vm based infrastructure.

N

They have like to have your model, which is like a hybrid on unified in infrastructure model, so our team is most of the time working with the customers on bare metal and will be also working with the customers on unified and also hybrid models. So, but today's topic will be covering most of the topics related to bare metal. Actually, so the first thing so, as I said in the previous slide, we took kubernetes as a de facto authorization engine by the time.

N

It's a very famous, not efficient engine, mainly satyr this project, like three years before when he started kubernetes. So there are multiple key challenges which we saw in kubernetes, starting if you want to have your ported cohort, cable segment, customers. The first thing they told us like when we want to run your Venus application attach to support multiple network interface. Suppose currently, kubernetes is doesn't support it and there are a lot of use.

N

Cases become as related to performance and also I availability of networking and how you going to do the cpu, pinning and Numa one else. Elementary there a lot of issues as we have. So we did a lot of brainstorming with talking with the customers and also with open source community, and we came up with idea of grouping them, so we group them. So this is something like the multi networking is goes to the kubernetes networking group.

N

How we going to ask you later: it's a packet, we developed a data plane, acceleration plugins and we put all the resource allocation and management things as the EP a future. So Hoover is so familiar with OpenStack or they might have here the term EPS. It's called enhance platform avenues so that you can manage your resources by service authorization was through the resource orchestration and the last one is the telemetry, which is also the important topic to discuss about service assurance.

N

So then, we developed multiple plugins in collaboration with open source partners and read, add basically and we developed for Mussina something we developed a multi data plane acceleration we developed with the use of space CNI and as far we say in a and bouncing a nice, and today we will be discussing basically on the networking side, so I will be covering the topics on multi-user space via nice, Authority and also the authority Wexler Ginn. So if you guys have any questions, you can interrupt me and ask any questions actually. So, yes.

O

This mean there are, there is there's a different CNI I have to use if I want accelerated data plane and if I want multiple network interfaces. I have to use a different CNI.

N

So the solution.

F

N

We provide is kind of a pluggable motor, so in this case, if you want to use user space again I, you want to use BGP or obvious switches for this case. It's that currently kubernetes doesn't support user space networking. What was in this case you have to use poultice so with, along with multiple, you have to use these space CNI, so it's kind of pluggable model actually most of the models which we discussed here is that answer your question.

O

N

So, let's jump into the multiple network interface for so, as I said in the previous slide, the problem statement is kubernetes doesn't support. Multiple interface is supposed to only one single interface in case of venous use cases you have should have at atomic level. Even if you ask you venus winter to develop your be enough application, they will say a Tomic level or they have to develop the arena. Fabrication which activities require two interfaces. Phone is just a fourth data planar control plane.

N

So they told us look we we need this future in kubernetes, the otherwise it's very difficult for us to develop a venous application and run it in cooperative. So the first project which we developed within Intel this month and the use cases as I said, is for the functional separation of both control, plane and data planes, and it also provides the different network service level agreement and basically one of the escape minutes. We talked with the customers, they said: oh okay, they don't want both control plane and data.

N

Plane should be in the same interface because various security reason. So this project, we were started in 2016 and we are trying to projects and we find the high data so not close partner along with us on this journey for our own wandered off years on pushing this one in kubernetes communities. These speeds started with the developing native kubernetes support of multi networking and then because of lot of effort from Red Hat, basically, some Dan Williams and Feng Feng team.

N

We could able to create a working group which is called kubernetes networking working group, which is basically focused on creating a de facto standard called multiple and cooking kubernetes. So in this working group is that is not closely on the implementation part. It's basically on. Oh, we can create a standard fit in kubernetes so that we can able to support multiple networking.

N

So we created a standard spec where we can define the network attachment definition so that basic network attachment definition is kind of aspects like a singing I spec, which you can implement in your CNI or in a device plug-in or in any one of your own controllers, based implementation.

N

The next one is user space say a nice, so, as I said in the my previous talks at the kubernetes, doesn't support or use space networking at all, so it basically works on the contour kernel networking. So for user space networking the major use cases the performance. So it's basically the software acceleration. So in this case we developed a plug-in which is called user space CNI with Red Hat, basically from belly, is on of the leading engineer.

N

Andy is leading this particular user space, see and I, along with other developers from Nokia and the Intel, and we developed as this user space CNI and it's the project is still in development phase. We are working along with various partner to make sure that is going to be a stable release and we're also planning to make it as kind of 20 ways pluggable motor. So we will discuss about this in the upcoming flight. What is the device plug-in and how we could help our networking actually.

N

The next one which we are working right now is SRB CNI slugging. So the problem statement in this case is kubernetes, doesn't support SRB networking interface, so I thought me. Network is basically kind of a hardware acceleration, so in kubernetes it doesn't support physical platform resource isolation. There is no guarantee to network I/o performance, so if you use Authority or Coover similar with sob knows that there's always a standard. The PCI spec, where you can divide your peer for the physical function into multiple gear, and each of us will have a guaranteed bandwidth allocation.

N

So basically, this saw CNI. If you look at the figure he's a severe which we, it pulls the VF from the horse network namespace as it moves to the cut in a network names. It's a very simple concept and we are elaborated this concept to make it work with a deep wreaking. So duplicate is one of the technology which developed from Intel, which calls data plane development stage, basically its bypass the kernel so that the performance is really more.

N

So if you combine both Authority and irrigate the performance, nearly at the line rate actually for the network interface card.

N

So, as I said about the debates plugin, so what is the best plugin? So most of the people in this meeting may be familiar with that it is plug in there. You might have here's the world from Nvidia as they're, developing GPU, the way plug-in and expediate a base plugin, but how device klingon could help us on the networking, because both are a different concept but or we could help us so before jumping into let's understand what is the base? Plugin means.

N

So whenever it for interview like click on, so we have various the devices like qat cards, Martinique and storage cards as well. So these are like a device specific cards. If you want to write all the things in cumin, it is nearly impossible stone, tsukuba natives, so this was management working who came up with a wonderful concept of develop device, plug-in concept.

N

Where you can manage these devices, you can advertise these devices kubernetes and the kubernetes could be able to keep your bookkeeping of these resources, and this also helps you to to the scheduling and selecting of those devices. But there is quite a few topics in this design is that it doesn't do that allocation part. But if you are having a networking resources, you have to allocate these resources when the pod gets filled. Only you have to do the relocation as well. So we came up with idea.

N

Okay, we have the authority device CNA, but the tropics, which, as service Ani, is that there is no mechanism we can expose, how many we have out there. In an oak, the kubernetes, scheduler and kubernetes ajudar have no idea whether you have a sort of iha nicking, your server or not. If you have multi clusters and you have to support like thousand nodes, it's very difficult to have a solution basis of an S or BC. Annex and and one more thing is the Numa awareness, the Numa awareness.

N

There is no guarantees that both the CPU and the NIC will be in the same socket. So there is no Numa awareness in this mechanism, so we came up with a solution that will combine both the device plug in and the CNI. So there are multiple solutions in the open-source community that they want to.

F

N

On the early on the device plug-in and the few people want to do networking on the CNI, but we thought that both these solution should work together, because kubernetes design at the end of the day supply couple models and it should work together. So in this case we Double, authoritative a plugin along with attacks, and this particular device plug-in acts like an intelligence which managed or knows the SRP resources in inside the server and Easter advertised those resources to the kubernetes and it allocate those resources.

N

So a sort of a device blipping is kind of intelligent guy. Who knows that what to do and when to do and the CNA is the guy who just gets the PCA information from the device clicking? In this case, we using monsters actually in this case some others just to get the information of the PC address from the cubelets and it passed this information to the CNA. So the DNA, the guy just, do the plumbing work ease. Does it have any intelligence he at all?

N

So this combination is really working very well and we see is the kind of a monster combination, bad for both the device plug in and the TN I can't go excess together and we are working closely with Reddit and sangseung team on this.

N

So all these projects is in close collaboration, which thing and we have a technical lead from for each project, something teams who we have to move and dance for the mulches project and for user-facing energy have Billy, who is the technical lead and for the authority based plugging the ABS Ronnie, who is working closely with another engineer called Abdul. So all these product is a close collaboration between the Intel and that so.

B

Crawler there is a question that came in from the chat which Arkady from Dell could ask our cardi here. Welcome to.

P

So the question I had was I mean you have a separate plugin for SRV.

P

What about other offloads, which need can provide? And more generally you know, the smarter needs which have multiple different offload now, how you gonna scale this model of the separate plugin for device capability.

N

You mean the device means like.

P

Let's stick to Nick's nuclear.

M

P

To Nick's you have a smart new cat, has a capability for the offload or you.

N

P

Is just one of the several possible thing that kind of float yeah.

N

So we are working with reddit on this one. Be developing. Martin exhibit plugging this kind of work of doing offloading with obeah switches. So it's still in development phase. So once we open source design proposal, you will have more information on average working. So it's basically your version of s or every device plug in right. It's work so long with user space so see and I actually for the offloading of fear for VR with virtual switches. But we are currently working on the mast.

N

Excuse case actually so into kind of a separate device plug-in actually yeah.

P

I would expect that you'll want to transition to that one. So we don't have a separate. You know plugin for each of the functionality of the offload which might exist on the neck. Otherwise,.

G

P

P

Different combination of the plugins.

N

So I think that could be. That could be one. You want to be able to do many things so I think yeah.

P

So I again I think there should be some level of discovery. What are the device capability with respect to the offloads and- and you know, obviously the scheduler should be able to understand. You know you know how a device device on the node. We should have that you know one of those capabilities and then I you know I would expect that the device have to be properly configured.

P

Violet might have ability to do multiple different of loads. I think it will be configured for whatever specific or flaws. You would like to do ah no more RN and that will be part of the configuration of the Nocona nodes and plugins will have to comprehend that those are available offloads for it versus.

D

Generic of both capabilities.

P

On the neck, yes, we.

D

Are we are working on no future discovery, which does what you described right, which is? We should be able to go to a host a particular hardware host and to feature discovery and label the nodes appropriately and run different device plugins as needed, but that's kind of the long-term vision that we have as well. Yeah.

P

But good, thank you.

N

Thank you. So this slide is just to say what is the transition from the CNI model to the CNA dBase plug-in model. So in this case you can see the CNA as to mechanism to schedule or consumables, like sov VF or no Numa alignment, so our team is also working on the pneuma manager. So in kubernetes we are planning on the next release with kubernetes, so we will have a human manager.

N

So the task for the starting point of this project will be like a showcase of the Numa manager can work along with CM k, CM k-means, CP manager for kubernetes and along with authority based plugin. We showcase that particular puce to both resource management, working group and also to the kubernetes plumbing working group. So the process, the development is still ongoing. So by next release, where we are more confident that will be along with the kubernetes.

N

So that is so I don't think. I am overshoot that time. Oh no.

C

He did quite well thank you very, very much and I sort of understand.

C

They're very good over and looking forward to hearing future stuff, and maybe Arkady and others can give presentations at upcoming events and are there any other questions? Did we ever them all.

C

Going and going gone all right, then I'm gonna have a czar step up and a little bit about this thing called edge. So.

H

Are you able to see my slides absolutely looks and is the audio okay? It's perfect. All right sounds good, so thank you very much Diane and team, and so what we'll do it in this session? The next 20 minutes or so is go through what does edge mean and what this will contain. This means within the context of edge. So first of all some definitions, you know, we've seen various definitions of edge edge is certainly the next infrastructure paradigm for delivering applications. Services closer to the subscriber and edge allows efficient, processing and delivery of time.

H

Sensitive data people have used different other definitions as well, which is some geographic distribution of compute nodes, distributed computing paradigms largely performed in the distributor devices, etcetera, but edges independent edges elastic as edges massive scale. It must be automated to meet that scale characteristic and those requirements are resilience and be built using public, private or hybrid. You know footprints with respect to cloud now, there's a lot to be said on just one slide in terms of what is edge. What does edge mean one way to look at it? Is it's it's about building.

H

You know ultra reliable experiences for people and you know objects when it matters and where it matters. So then you know many times: people struggle with edges, customer edges, provider, edges you know somewhere in between. It can be anything that sits between the subscriber and the provider cord or provider data center. And if you take that our definition of edge then automatically it will, you know you'll start to look at. How do you solve different application challenges? How do you solve different deployment challenges?

H

How do you solve different Network challenges in the context of edge, because then the footprints will vary, the sizes will vary. The the deployment matters and context will vary. So, let's, let's start there now we know in a telco space. Typically provider network can be classified as sub classified into many different tiers.

H

Not all these tiers apply to every single provider in the world because geographically, some of these tiers are merged, and you know in in smaller countries and in in countries like you know, US and Canada in Australia. You might have actually you know much more enhanced set of tears, but certainly there is you know the the access network access a delegation network there's a termination infrastructure.

H

There is, you know the last mile copper fixed wireless in terms of the wireless side of the business or or the copper cable Jeep on optical stuff, and then there is the customer CPE, whether it's a residential CPE or a business CPE, and then the actual customer network itself. If you look at that tearing or that categorization model I mean this is a work that was done by between me and fuzzy and credit to full credit to posi for actually coming up with this kind of a tiering model.

H

And you know it then allows us to segment what is applicable in each one of these tiers and then once you understand what is applicable in each one of these tiers, then you can understand the type of applications. The type of network functions, the type of capabilities that are required at those tiers and then try then try to solve the problem of what is required from a you know: infrastructure perspective. What is required from a networking perspective, what is required from a containers perspective and solved so on and so forth?

H

So you can keep double-clicking into each level and then define the level of detailed set of requirements. I mean everybody talks about edge, I mean here's an example of a slide that was done by China mobile at the vco keynote presentation in Amsterdam, and if you look at this in terms of the deployment model from national district level, data center down to the base station- and this was because this is China Mobile- it's talking about Radia infrastructure. The number of servers that they need to deploy this whole virtualized ran with edge.

H

Compute is about 1.25 million service in terms of number of sites number of locations and then multiply each location by the total number that they need in about 20-25. This is their estimates and the distances that these reside from the base station or from the user and the latency requirements in all of those I mean we ended up.

H

Actually, you know again credit to my friend Fozzie here for actually coming up and- and you know taking pains to catalogue all of these details- and the important point here is what I really want to show through this slide is not necessarily to dig up. Why or what, but to tell you that, regardless of which way you cut it, it's gonna, be large I mean if you take this number and apply it to a number of cell sites, a number of locations or number of points for AT&T or Verizon here.

H

That still is a pretty large number. That still is, you know two hundred thousand three hundred thousand whatever that number is here. It turned out to be more than 1.25 million in terms of the total number of servers that are required. But what's important in that context of scale is when you have those locations deployed with compute infrastructure, then functionality from the core data center. From the you know, regional data center starts to migrate towards those locations.

H

When that functionalities trust starts to migrate, because you are providing better experiences. Remember we started with the definition of you know edges about providing better experiences when you start to provide those better experiences. When that functionality starts to migrate towards that edge, then you need at that edge infrastructure that edge location, all types of capabilities, ability to manage virtual machines, ability to manage containers probability to provide an assurance framework. You need a view of where those worth nodes are placed, how they are placed.

H

What's the your assurance framework, what's the overlay network in control or under the networking control for all of those? Now you just saw presentation from Intel a moment ago about you know what are some of the networking challenges specifically and how you know: Red, Hat and Intel are working together to solve some of those networking challenges there is you know more than that also networking challenges. We also need things like visibility into the nodes. In terms of how do you place functionality, how do you distribute functionality?

H

How do you manage storage in in in those contexts? So this is a a massive challenge. Scale brings a massive challenge with it, and with that you know the workloads, the different types of workloads, whether it's virtual machines or containers, need to be able to address all of those things.

H

It's computing generally service category scenarios, if you actually broadly look at that, there are various different types in terms of how you can categorize or divide. There is the IOT application side. There was a network performance application side.

H

Is the enterprise connectivity side, this third party network, peering and hosting capability, this customer oriented residential mobile types of services or operator services, which is the deployment of base stations or VPNs or cash CBN type of deployment, and regardless of which one of these you take the number of locations and then multiply that by the size of those, you know requirement that you, each one of those locations, you're, really really talking a massive infrastructure.

H

Now, when you have that, when functionality starts to migrate, whether that functionality, that was in core datacenter running in kubernetes or running in OpenStack in virtual machines, now has to be able to run at that same edge location and be able to be been managed with single pane of glass across both of those locations so just putting out there in terms of the scale size and putting out in terms of the complexity in terms of how he managed. So that brings me to okay. So what are those use cases that are being deployed at edge?

H

What are people considering? So if you take a pure telco view, you know from a telco standpoint, a telco delivers enterprise residential or you know, mobile types of services, so those services are being terminated. Today on a central office, we have now a central office blueprint based on OpenStack that central office blueprint now needs to convert it to be a container blueprint as well with I mean instead of OpenStack or maybe both, because you'll now have some of those network functions.

H

Moving into containers, then, when you deploy just pure edge compute for other traditional applications, IT applications at the core is a dent. With these network functions, then those are running in micro services in containers, and so you need to be able to operate those in the mobile case. For example, you have this beer and virtualized ran now that virtualized ran is becoming containerized ran, so I mean you're.

H

Taking those you know, ran application and running them in containers and I have actually have a couple of slides later on to explain a little bit on that in terms of what's going on there.

H

Then there are other interesting areas such as video, optimization that needs to happen that needs to happen locally. In that edge, complete location, then you have specialized deployment models such as stadiums or many use or big public venues, and then the technology itself is moving from an energy perspective to containers and we'll actually look at some of that. Some of that in a little bit of detail, you can also take things like in the IOT space.

H

You know analytics is a thing that keeps coming up every every time when we talk about IOT and H compute deployment is almost necessary to provide timely feedback into those sensors and devices that are autonomous that are operating independently. Otherwise is the round-trip delay times and then the complete processing time at a centralized data centers not sufficient. You know in terms of how quickly you need to take action in those type of situations. Then there are traditional telco services, which is your specific. You know, CPE managed, CPE vnf as a service and so on.

H

Today, some of these are delivered using dedicated devices, MANET CB stuff tomorrow or pretty quickly, moving to a virtualized model and pretty quickly will move to a containerized model as well. So the what what you will see is again more in the word use this extensive thinking, that's being put together by telcos to say how should I go about transforming my existing set of services and into this new model, and how do I go about deploying those again?

H

We can talk about each one of these line items for a very long time, but what I want to give you through this particular list is a mix. The mix that's being deployed at each compute layer is both a set of network functions and a set of traditional applications. The idea plication that will run on edge, which means then, whatever was typically running in a centralized data center, now needs to be running at the edge, whether that applies to virtual machine context or whether the or it also applies to the container context.

H

Now we'll just take a quick drill down into two areas: 5g and 5g, because somebody brought up 5g here on the topic, I think it's relevant to take a quick drill down 5g seen as an enabler technology for new use cases. It's also seen as a leveling the playing field between incumbents and new entrants into the market. There's a lot of you wouldn't believe that there are a lot of providers, people around the word, speculating jumping into the 5g spectrum auction and and getting into that field, because it's it seems attractive to them. Why?

H

Because it resolves some of the life last mile access, it provides an instant you know, subscriber access. If you drop a high-speed modem into a user into a residence or into an enterprise customer, you don't need to worry about whether that user has a DSL or a fiber or a cable or or any other connection. You can get potentially a high-speed connection into the user by just dropping in a modem and in their house or in their premises. So it results some of those last mile challenges and that's why fixed mobile broadband?

H

That's the term that people use. It is a very attractive service option based on 5g for a lot of providers. Then, of course, because of the low latency capability. Because of you know massive scale and machine type communications. New services are possible through these enhanced use cases for 5g, you know, and then some of the deployment models lend us to deploy better H compute capability with 5g and which means again better user experiences.

H

Overall, however, one thing that's true, though, is it does require huge investment in terms of the infrastructure in the band space and from a mass scale perspective. We probably still you know, 18 24 months away. Phones are beginning to be announced or user equipment as being beginning to be announced.

M

H

With availability sometime later next year, so then you can see some spotty deployments coming up. How long did it take for transition from 3G to LTE? Think about that? How long will it take us to transition from LTE to 5g? We can all speculate in terms of how much it takes and how much investment they'll do again.

H

Just because this is a use case. Centric I wanted to kind of point out that you can take these specialized use cases. Applications database requirements all of those things and categorize them into these three broad categories of 5g, with enhanced mobile broadband, machine-type communication or low latency communication, and then you will see that how you know each application service or each application Falls can be categorized into my own one of these three buckets and then deployed from a telco perspective.

H

I don't want to go into the details of you know how we ran works and how here and work for LTE or 5g. But what I want to do is give you a pictorial view of where that compute nodes are deployed in the infrastructure and what is being becoming a candidate for containerization.

H

Going forward so if you take the 5g architecture, you're splitting the functionality that was in base man unit, probably at running as a virtualized base man unit into what we call a centralized and distributed nodes, these centralized and distributed nodes are certainly candidates where you can run these containers. You can run these code in containers today they can see, use their virtual machines in terms of any other trials and any other conversations we've been having internally.

H

We showed that at the open networking summit in Amsterdam as a demo POC, but you know there is a lot of talk and there's a lot of conversation around how this implementations can move two containers pretty fast, not only that, with these infrastructure components moving to complete containers, there is this standard compute nodes that are being deployed right next to those locations. What we highlight in this picture as edge compute right here inside these turrets red circles, that each compute is your generic. You know computer infrastructure running your traditional applications. It could be.

H

You know a caching node. It could be your traditional. You know localized database. It could be your traditional IT application, which would be something that does data analytics with the data. That's coming back on this particular infrastructure right there locally, rather than at a core data center, especially that edge compute node, allows you to now shunt locally traffic with with 5g, where you place the user plain function over there, and then you can do a whole lot of things with different kinds of services and experiment with that.

H

So this, this compute node infrastructure can will have to then run any application, which means containers, VMs alike right and those things translate to now many different requirements on our site. So this is about the the ran side.

H

If you look at the packet core for, for you know, mobile packet core has evolved from physical devices in the top left side corner to virtualized capability being deployed and OpenStack, which is widely done today to the new paradigm, which is being discussed right now in terms of or of being deployed right now, which is cups, which is controlled data, plane separation for better flexibility to the top right corner to the ng core specification, which is the next generation core four 5g, which is potentially all containerized and all of these functions run in containers and the very next slide.

H

Actually, let's skip this in this slide gives you a picture. This is courtesy, affirmed, networks.

H

This is their slide and when you know they presented, this I said imagine that we are taking these 5g core functions and container izing them building a microservices model and here's how we are thinking about refactoring those there is a set of core components of core, or there is a set of common micro services that we will use across many different components inside our network functions, fault management, logging, HTTP licensing security registry right in memory, replication, MongoDB, activemq, different types of plugins for network like calico plugin.

H

So these are micro services that are already available as part of a commercial catalog. These micro services are common too, then, then all they need to worry about is how do you leverage these micro services and then how do you build the core functionality? That's required by. You know the 5g components like access management function or session management function, which is am FSF's as the service-specific micro-services, and then they can have a set of common networking, routing services and protocol or balancing micro services as well. That leverages this.

H

So now what was previously packaged all together as a single virtual machine or a set of merchant machines, as part of me and F, is now completely refactored and broken into all of these micro services, and then these micro services need to be managed as a cluster as a group and then deployed in these various different thousands of locations that that we need to do so. That brings me to my you know the last few slides, which is so container izing.

H

These things puts in a lot of requirements a lot of requirements on you know the infrastructure, whether it's so kubernetes as a container management capability, needs to be aware of certain things.

H

Then how do you bring about the the other functionality through custom resources through CR leads to custom controllers through service mesh and network mesh through you know what other things that we need to be able to add to bring about the similar functionality, then you've become a little bit used to in in the VM world at the moment of VM, or you know, virtual machine world and bring that now to the container.

H

So multi-site is is incredibly important with respect to you know edge, so provisioning of those multi sides provisioning of those remote nodes, the zero touch models, the single pane of glass of Orchestrator model, in terms of how do you place that functionality and where so stuff that was discussed earlier on the call by detail in terms of node feature discovery, what Frank just answered they become? Incredibly, these type of features become incredibly important identity and security management confederación capability, because now you have thousands of these sites.

H

They need to work together as a single cluster, you need to have a common identity across all of those clusters, common way of Orchestra managing those h.a models. Do we need a kubernetes master node on each on each one of those sides? Do we need some sort of a headless operation part model on on those sides, at least when you restart these containers, can you restart a single container? If you remember my previous slide in terms of how a single vnf got refactored into all of these different micro services?

H

Maybe you need to restart differently in terms of bundles in dependency trees? How do you create those dependency trees and do worry about restart ability of those application? Bundles multi-tenancy is incredibly important for telco in in often we ignore multi-tenancy, because we are typically looking at deploying containers inside a single enterprise.

H

We in the telco space now you're talking about taking a service that needs to be offered to thousands of subscribers thousands of businesses, and you cannot have an application that shares data for those different subscribers. So you have, you may have copies of applications running you know in different tenants. Then those tenants need to provide they get their own view of, what's going on from and from their application perspective from their infrastructure perspective.

H

Networking requirements, I think there's already being talked about enough, there's more in the context of not just Malthus, but also things like native v6 native v6, everywhere know. Nehring know you know v4 to v6 conversions and that becomes really really important. Then data plane acceleration compute in terms of real-time capabilities, thread, painting, CPU, painting models, I think Intel did go through some of those things that are coming down. How do you patch OSS?

H

Can you actually worried about the dealing with this as a black box approach versus an incremental Apache model? If you're, if you have 1 million nodes, how can you do incremental patch, across those 1 million nodes and and and on and on and on right I mean I can keep going on with respect to these set of requirements? Some of these things are being worked. Some of these things are yet to be work, I, think through this community.

H

What I'm interested in is collaborating working, defining refining these set of things so that not only when you know different groups of people are involved with this, get the right set of things to go then build the develop within the community level and at Red Hat level and and participate in that. So that's kind of a quick summary of where things are with respect to edge in containers.

H

Again, it was to provide a quick high-level view, and it's about time as well.

C

Well, that's great: is there any questions right now with someone piping in there with a piranha.

N

Coral actually to in one of the slide, you mentioned a good ipv6 right, so in this use case, you're looking for a kind of peer, dual tech support for both ipv4 and ipv6 in kubernetes, actually.

H

Actually, both people are looking for dual stack as well as people are looking for just native v6 across the board, no dual stack. So if you go to Japan- and you want to talk to any of the telcos in Japan they'll say give me everything native v6 today, there's no more v4 addresses being allocated by Ayana done so.

N

They did thing only near ipv6 yep, some.

H

Of the cable codes are also saying the same thing. One of the cable coil talking to big large cable go right, they're saying no! No, we just don't want to bother about this anymore.

N

H

We all know Natick models too. That's the other thing.

N

Okay, because we've seen one of the customers asking for boo in same in PI G use case to both ipv4 and ipv6. That's what I was thinking like when you say an ATO, ipv6 I've also just understand all this.

N

So much looking forward I mean knowing the phone customer say they will say IP the phone and of both ipv6 I. So that's what I want to actually understand the value industry is getting yep.

C

There are any other questions you know. We've run over time a little bit, but I really wanted to thank you for for giving the presentation it really kind of helped set the stage for all of the other topics that we're going to need to talk about over the next year or so as things develop. I have proposed the next meeting date to be the 25th of January aka. The fourth Friday of each month at 8 a.m. and I haven't heard any objection so I'm going to set that up.

C

I know some of you are in act pack, time zones, so that might not be perfect for you, so we will try and determine in the future, and there's been a few other folks chatting in the background too. So, hopefully we can get, maybe a contrail. You know, Paul latzke Lancaster was working off with another I think it was my OC or my something or other another telco service provider to get them on on board.

C

For the next meeting to talk and I'm, hoping that this structure works for us, I tend to like to have these like at least two little talks to spur conversation. But if we need to only do one to talk, and so we can have more conversation, we can even shorten that up in the future. So if anyone has any feedback, I've put my email in the chat, so you can send feedback that way and also if anyone is come in to coop con.

C

Please pay me because I will give you all three passes through the gathering, which is on December 10th and, as I mentioned as well on the chat. There's a very large exhibit room lunch, slash lunch room next to the room where all the presentations are going on, and it would be a great place for everybody to maybe have lunch to get or sit together during one of the cotton breaks or in the reception.

C

The just connect face-to-face and then have a sort of face-to-face informal state meeting on December 10th there anything that anyone would like to add to this conversation today. I know we've run over time. So actually both had to drop off, but then played a good introduction, I think through both the people in the space at red hat and a lot of the partners here and a few are the customers too.

C

As our, if you could email me or add a link into your into the notes to your slide, deck I will publish all of that. I will try and coerce Paul Lancaster into doing a recap: blog post.

H

Send you the link? Okay,.

C

Perfect all right guys and gals, and everybody else please join us next month on the fourth Friday at 8 a.m. and I will send a notice out and try and build a landing page for the telco saying on open ship Commons between now and then so have a great holiday. If I don't see you in Seattle.

H

That's good thanks.

H

A

P

O