Cloud Native Computing Foundation KubeCon + CloudNativeCon Europe 2020 - Virtual, 4 Sep 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: KubeEdge: Kubernetes Native Edge Computing Framework (Intro) - Yin Ding, FutureWei

Description

Don’t miss out! Join us at our upcoming events: EnvoyCon Virtual on October 15 and KubeCon + CloudNativeCon North America 2020 Virtual from November 17-20. Learn more at https://kubecon.io. The conferences feature presentations from developers and end users of Kubernetes, Prometheus, Envoy, and all of the other CNCF-hosted projects.

KubeEdge: Kubernetes Native Edge Computing Framework (Intro) - Yin Ding, FutureWei

KubeEdge is an open source edge computing framework that extends the power of kubernetes from central cloud to edge. It provides application management and service communication across cloud and edge sites, device management for multiple IoT/Edge device communication protocols with lightweight implementation. During this talk, Kevin and Jie will review KubeEdge motivation, architecture; then go through latest updates on new features and user adoptions. After that Kevin and Jie will introduce where the project is heading to and how new contributors to get involved. There will be an open Q&A for attendees to ask questions.

https://sched.co/ZevB

A

Hi everyone thank you to coming to my session. The session is about the kuber edge. The name is uh the kuber edge, kubernetes native ad cloud computing framework. This is the introduction session. uh My name is indian. I work for future way. I'm the one, the founding member of this project.

A

So for my talk, I have this structure. I will talk about the background, then a little bit architecture of a kuber edge. Then the community status following the slides I'm going to give a user adoption example, and I also going to show a a short demo to see how easy you can deploy your own core badge for your use and at last I have a few useful questions: information to share with you and I'm going to do the live q a after this recording.

A

So uh please have your question ready, I'm going to live to answer your questions online and or in person. If I can so, let's start uh background, so kube edge cool edge is a platform we build upon kubernetes.

A

So the purpose is to provide the fundamental infrastructure to support network application, development and metadata synchronization between the edge and the cloud. So all we know cloud and edge are very different. So there's a lot of challenges if you want to resolve edge computing problems. First, is network reliability uh not like a cloud. You have a high speed, reliable connection uh within the servers within your data center, so cloud to edge. You may have unreliable network connections and your network bandwidth will be limited between your cloud and edge and also a lot of time.

A

The ads have only limited resource, so we need to work within that limitation and also uh when the network got disconnected.

A

We need uh autonomous operations on the edge, so the the edge can run independently without connect to that cloud, and also because we support a lot of iot scenarios, that's a highly distributed and heterogeneous device management and iot protocols. We need to work with. So that's all the challenges so uh with this challenge, coverage can provide like uh four major features. The first is seamless cloud edge communication, so we do this for metadata and data seamlessly without the user, interaction or intervention and the uh we also support edge autonomy.

A

So, basically, we support autonomous operation on the edge, even when the edge is disconnected from the cloud and also we are low resource ready. That means, if you have only a low, uh config, the edge server, uh the combat can still work in this resource constraint the edge in this situation. Basically, you have a low memory, a low network bandwidth or a low. A very low end, cpu and low compute power kubej can work in this scenario and also uh we have a strong support for the iot.

A

So we simplify the device communication. So uh it's very easy for the iot application to communicate between the application device in the iot scenario.

A

So this is our release journey this project we entered as a sandbox project of kubernetes cncf in march 2019. At that time we already support kubernetes core apis, including a node, the pod config map, secret etc. So in the june 2019 we have the official 1.0 release release.

A

During this release. We highlight some features like edge cloud coordination for devices and also we verify the container integration.

A

Then, in 1.1 release we provide a csi integration for the edge and a 1.2 release. We uh a big feature. We release is kubej component config in the 1.3 release. We support a log connect from pod at edge. We also verify the crio integration in 1.4 release.

A

That's in the in august 2020 uh we are going to have a uh we're going to support kubernetes a kubernetes native matrix api, so you can collect performance data from edge to cloud and there's a few more features, because when I record this video we haven't have that release. Yet I hope, when this recording is, played we're going to have this release, so I gave the release log link here, so you can click through when you see this video.

A

So now we are released every three months then about one month after official kubernetes released, we are going to have the compatibility issue to solve. Then we can up upgrade the kubernetes dependency, follow the release. So we are in the same three months: release cycle now.

A

Architecture, so let me introduce the kobach architecture.

A

Basically, uh cooper edge is natively extended kubernetes, so we work with kubernetes we may in the majority we have two components cloud and edge, so I'm going to elaborate each part in the following slides, but now I will you introduce so mainly on the cloud. We have a cloud core. Basically, it's talk to kubernetes and proxy.

A

The request to the edge part, the edge, have a edge core and it's uh talked to the container engine and also the amateurity devices, and I'm going to have this elaborated in the following slides for the uh edge cloud part. So basically, we have the least watch to talk to the kubernetes master. We have edge controller, basically, the shadow, the management for the know, the part, everything uh kubernetes native resources and we have a device api device controller. That's for the iot devices modeling.

A

Then we have a sync server. That sync server is basically uh we can detect inconsistent between the cloud and edge and to solve these issues, then we have the kubernetes csi drivers does uh who have the storage provisioning for the edge, and also we have the emission web hook. That's for the device crd. We can do the api validation and the best practice enforcement edge, so one of the important part is edge so in edge we have edge hub.

A

Let's talk to the cloud, so the cloud part of kubernetes have also a cloud hub, so that's the we set up our websocket from edge to cloud, so we can talk seamlessly uh from edge to cloud. So this way, even your edges behind some firewall gateway. When you don't have a public ip, we can initiate a connection from azure cloud. So then the edge cloud can talk through this websocket channel to to reach the cloud edge. Communication and the other part we have a metadata manager is a load level of metadata.

A

Persistent and the hd is more like a cubelet, but it's lightweighted and only do uh necessary jobs, and also it's uh also responsible for the power of the lifecycle, and we have a device too and that's basically, a synchronized device status between the cloud and edge and the device.

A

The last part is the event bus that is, our immigration. Client does uh synchronize the device status from the device and the edge node. So to in this way, you can control your iot device.

A

Community, as I said, this project entered cncf sandbox as a sunpower project in march 2019.

A

So from now from that time we grow uh very healthy, and now we have more than from 30 to 300 contributors and the github starts from 400. Now it's more than 2700 and for forks github forks at that time is 100. Now it's more than 700 and the contribution organization is from beginning. Only one now is more than 25 and maintainers is growing. Healthy uh from seven to fourteen is from five different companies.

A

uh You can see the top contributors, the new contributors are from four different companies, and then we have a steady com, commit trend for the uh our ripple and also we are collaborating with other open source communities and we, the kubernetes iot edge. What group working group very support for this project and we have a very close collaboration with them. Also, we work with uh lf edge linux foundation, edge acrino project. We have a two blueprint project family there. One is called elliott, it's a mini for industriality, the other one is kuber edge and service.

A

Bp is a blueprint. That's for ml offloading and some ai framework, a lot of new verticals or scenarios will be built.

A

Contributions, so you can, as I said, we have a lot of contribution organizations and also we have a lot of adopters contributions, including arm samsung, china, unicom, etc, and that's all listed here and also we have more than 20 adopters. uh You can click, that's all listed in our github repo. You can see, we have adopters a list and we mentioned the risk com, work, cloud, xinhi, ot, coop sphere, huawei cloud and a few others. We need their confirmation we can before we can publish their name.

A

So one user adoption example I'm going to present here is a highway, etc system uh you can see. uh Does uh we have a lot of a tow gate in the highway? So does a touchless contact list, so the the car can drive through the gate, so it automatically recognizes your uh play number and charge on your account, so you can pass through the gate quickly without a stop and go and pay and keep you changed or something so. For this scenario, we have a lot of edge.

A

So basically also the ads have a heterogeneous hardware. They some of have 86 servers. Some are using arm servers in their gate of them. We are our edge. Node is controlled by the kubernetes plus kubernetes, so kubej.

A

In charge of deploy the application to the edge and manage the life cycle, uh all this stuff, uh the uh kubernetes and code badge is deployed in the public cloud and edge is uh geo geo geo distributed across a lot of uh places.

A

So in this scenario, uh we have more than fifty thousand, as you know, managed by the kuber edge and also the containers, the application containers in more than half a million so and also it will it's generated about 300 million data record every day and it's improved the efficiency of this system. It used to be a 15, a 15 second, to uh process a car. Now it's only take two seconds: uh the truck used to take almost 30 seconds. Now, it's down to three seconds: it's improved the user efficiency.

A

uh Now let me uh give me a few minutes: I'm going to show you a demo to see how easy you can deploy your own kubernetes system and how is easy, how easy you can use the same kubernetes experience to deploy application from cloud to edge, so you may not even notice it's running on the edge. It's not in their data center.

A

Now let me show you a quick.

A

A

So basically, I have a two windows. uh The the upper one is the edge the lower one is the uh cloud we already have a kubernetes deployed. You can see, uh there's a only master node right now. Now we register our cloud part. It's just one single uh single command. You can see uh the cloud is registered.

A

The cloud part is the cloud core we I show in the first slide of the architecture. Now we keep a token. This token is using by the edge to register to the system.

A

Now we uh re initiate the connection from the edge of the cloud to register this edge node to the system. We use this token and to verify your identity.

A

Now we can go back to, we can see the log that the edge core part is running properly.

A

Now, let's go back to the cloud: let's you can see in the screen, you the get node. You can see the the agent the edge node already registered as a worker node in the system.

A

Now, let's do a quick deployment. Basically, we have a simple ngx.

A

Application we are going to deploy, so you can see all the api and the cli is the same as the.

A

It's all at the.

A

Kubernetes now we you see that we just do the deployment uh yamo is deployed. You can see on the top. Your id docker ps, see the system I see deposit running now and at the bottom we from the cloud we access that njx port. You can see it's responding. The njx is running properly.

A

Yep, that's the quick demo.

A

uh So you can see from the demo it's just a one command. You can regist the. uh Let me go back to.

A

My architecture, so you can see just one command. You can register your cloud core connect to the kubernetes, then, as just one command running on the edge you set up the connection from azure cloud then register in the kubernetes.

A

uh From that time you can use the native uh the same kubernetes api to deploy the application from the cloud to the edge. Even the ads is running behind some firewall. It's normally even don't have a public.

A

Ip, so so in the future we are going to. uh Are we in a good uh trend, so we are going to build our community uh even better, so we welcome everybody, contribute and participate in our community. We are going to achieve. We are going to seek more wider user adoptions and we are welcome and hope more developers can participate and uh for the scenario we are going to have set up a two things. One is a device: iot one is a mac, it's a mobile edge computing or a multi-access cloud computing.

A

So we hope uh with this 2-6 we have more user cases and more user adoptions in this project, and we are currently applying uh graduated from the sandbox stage into an incubation stage stage. Now it's in the middle of a tcoc voting right now, hopefully, uh by the time we play this video, we already uh formally be enter the incubation stage of the csf incubation state uh for more technical.

A

We have a detailed project uh developing the roadmap. I will come, everybody go to our a uh github to see our roadmap and to give your feedback. If you want to uh new features uh you can put in here or issue a pr in the github, so uh this page, I'm going to give you are here, is a the first one is our of official website kubernetes.io and the second is our code repo, so you're welcome to get go there and issue your pr to re. Do the feature, requests or bug report and any questions.

A

You can also go to slack uh you all this. uh You can. I believe this slide will going to be shared on the website. You can come here and click here to join our slack channel and to talk uh to have your questions or feature requests. Anything, and also we have a google group, so you can join our mailing list. Ask questions as well and we have a regularly uh meeting basically by weekly. One is a europe friendly time. One is north america friendly time.

A

You can pick up your more convenient to you all. This is in our github repo and also in this slide you can click through and then we have our calendars. You can click in all this meeting. uh According to our csf policy, we are recorded and published in the youtube, so you can go to our youtube channel to see all the meeting records.

A

So if you have it so, if you miss one meeting, you can always go to youtube to see the recording and use the slack, mainly any channel, to ask our questions or send you a feature request.

A

So now is the time for q a. I will be here to answer any questions you have so now. I'm going to thank you, I'm waiting for user questions.

A

I got a few questions. Let me uh I just try to answer that in the order is okay. The first one is the communication between the edge and the cloud node ad hop and the cloud hub. One is how the edge node amount themselves.

A

So the question yeah we could put an net between edge hub and the cloud hub, so it doesn't matter it the goal. We do the websocket this way is to make sure even you don't have a net on your edge on your firewall for your edge. You still can communicate between the edge node and the uh control control plane and the. If you have a net that that also works and also from websocket, we have tls.

A

We can configure trs to do the secure communication and among the effect, if you have- or you don't have a net setup for between the edge node, we have a small project called edge mesh to solve. The problem between the communication between edge nodes is still in the first stage. We are going to improve it in the second half of this year.

A

If you are interested, you can go to the github, there's a edge match folder, you can take a look and also in the dock, there's a proposal for the edge mesh. Are you welcome to give some feedback on the proposal? We are going to update that proposal for a more better solution.

A

The second question is oh about the state of kobach yeah. Currently we are still applying gradate from sandbox it's in the middle of a uh tlc voting. We got the six votes already. uh We need eight to uh officially graduate from uh grand graduate from sandbox into the incubation, so hopefully, as uh in the next few days, uh technical companies. uh Actually we uh got a few uh technical companies. I they have a different purpose to use of akubat, for example in the china unicom.

A

They have their own their cloud, offering using the kube edge for their iot platform, all this stuff and for the mec, the teleco nfv stuff. We have a mec, uh sig uh special interest group in our project, so I'll discuss how we apply cobalt into the uh fv. All this mec scenario and uh a few already a few technical companies uh show the interest uh in the discussion.

A

However, uh they are not in the production yet, but they are evaluating and prototyping the solutions. Hopefully, uh in a few months, I can give you a more detailed update when they agree to disclose their names and their open source efforts.

A

The next question is uh ask about the edge. Node is the part of the cloud. Yes, from the point of view from control plane, you can see the edge node as a kind of a worker node. The only difference is uh the etch node at the remote instead of in the same data center.

A

So the as I said in the presentation, the node, because it's in the remote you probably connected through the isp internet, so the connection may not be reliable and it could and it could be, a slow. The latency could be large, but the from control plane. Point of view is part of a couple system.

A

So it's a just a kind of a work. Note you talk to a worker node uh the same way you talk to the regular worker node. However, the kubernetes will proxy your uh all the communication between the remote edge node and the control plane.

A

The next question is about cool badge.

A

Why should I use kubernetes instead of a mini cube, so a mini cube is a full kubernetes deployment, so kubernetes won't ins uh replace mini cube because cool badge doesn't include the uh all the implementation of a kubernetes. It's extended it's built on kubernetes, so I probably the demo is very short.

A

The demo is showing actually it's using mini cube. We in the cloud node we deploy kubernetes so to simplify deployment. We use a mini cuber deployment, so the coupe edge is add on to the kubernetes. Basically add on to the mini cube, so kubernetes won't replace mini cube. Mini cube is a kind of a very uh simplified deployment for kubernetes, coupe edges that add on to the mini cube.

A

So if you have a remote edge node or you want to uh control remote edge node, you can use mini cube, plus kubota, as I show in the demo. So the google address the goal is to control and to manage the edge node at the remote site and to solve your network connection, issues, authentication or other deployment issues.

A

The regular kubernetes, including mini cube, won't solve this because they cannot tolerate the uh slow internet and unreliable internet connection and also it's maybe uh it could be the disconnect digital could be disconnected from the internet. You still need to handle the request.

A

uh Another question about the uh the demo: oh yeah, the demo- is that, because timeline limited, uh the question is uh why we show this kind of demo. Is the the njx port running on the original yeah? I don't have time, but I I can show you uh later is the njx port is running on the remote edge note, so the the demo is show two purpose.

A

uh First, one is very easy to deploy a uh kubernete kuber edge onto the existing corvette, kubernetes deployment, and the second purpose is say: oh yeah, after deployment that you can easily using the existing kubernetes api or coupe control the the the way it could exactly the same as kubernetes to deploy your workload to the remote edge site, and I didn't have time to show how we do the edge devices.

A

However, I can, if anyone interested I can, you can go to the our github. We have an example: folder there's a kind of example. If you have any questions, you can go to slack to ask and I'm happy to build more demo videos to show.

A

Let's skip the next one. The question you can ask here, I suggest, if you have a more technical question, asked in the slack that you get a more detailed answer. If you ask here, it gets lost after this session. I won't have access.

A

A

uh The another question is about the the cube quick and the websocket. We are evaluating the quick, maybe replace websocket. That's just our technique that we'll try to evaluate which one is better.

A

The last question about the: how what's the best developer edge application with kubernetes how to respond responsibilities between cloud and edge. So our principle is: it depends on on your job right, so the edge cloud computing, the the purpose. The goal is to compute and process data near the location that data is generated.

A

So that means, if you have a lot of iot devices, for example, you have a lot of iot devices.

A

You collect a lot of data so in order to process that you, so you don't want to transfer all this data back to your central cloud to process that it's a waste of bandwidth and there's the also the latency. The response latency will be long.

A

So that's why you want your processing job on the edge side, so you can have an instant result to process data, also the data privacy, for example. If you are a manufacturer factory, you don't want to upload your raw data back to the cloud in case. There's a security issues you don't want to.

A

I mean to expose your data to the public. So if you can mask the data on the edge, so then you can even there's something wrong with the cloud. You won't have issues so basically that's the two uh so to use so when you try to split your responsibility to the cloud is: if you have a preprocessing for the privacy, do it in the edge side? If you uh want a low latency uh and to then you do this on edge. Also, if you want to see the balance, do it on the edge.

A

However, if you have a large machine learning or a job, you have a energy consumption.

A

It will come some consume a lot of energy on the compute or you need a large computer resource. That job should run on the cloud, but you can add, as you do the low latency to save the bandwidth and also to save your privacy.

A

I think that's all.

A

Okay, that's a.

A

This uh last two question is sorry. That's the last two questions is: uh what's the updated component, uh that's one we're still working on that in the community. We have. Let me try to produce a better upgrade plans.

A

And the last question is about: can we manage a remote cluster node? Currently, no, we we have only one control plane. We compare uh if we want the remote cluster on the node, so so we evaluate the the scenarios so far. We think we treat the node individually. However, we are going to introduce edge mesh to allow the the node to communicate between each other, because if you allow the cluster in the node, you need to synchronize not only the node status but also status cluster status between the master class and small.

A

I mean the the remote cluster uh that one we evaluated, we think uh probably, uh for example, in the production example, I show in my presentation they use multi-clusters. However, it's more like a cube of fat to do the uh layer, the cluster in the cloud side, instead of in the edge side.

A

So that way uh we we don't want to. We have a trade out between the complexities and the and the simplicities. So so far we only control the remote node. However, you can allow the node to communicate each other for you from your application.

A

Okay, I think that's all the questions. I have today uh thank you, everyone thank you to attend to listen to this presentation and yeah. If you have more questions, please go to the google group and the slack I'm happy to answer more questions there and I hope everybody can. I thank everybody showed interest. I hope you can have a chance to try it and probably to participate in our community.

A

Thank you, yeah. Thank you. Everyone.