KubeEdge KubeEdge Demo and Usecases, 10 May 2021

Previous Meeting

⏯

youtube image

►

From YouTube: HDC2021-KubeEdge Explained: CNCF First Cloud Native Edge Platform

Description

KubeEdge is CNCF's first incubating project for cloud-native edge computing. Tune in to watch this session now and learn how to participate in the open-source KubeEdge project.

A

Let's begin this afternoon session, so this session is a session concerning cloud native and edge computing, and I welcome all of you.

A

Let's begin today's session, I'm the first speaker, I'm shife from huawei cloud. Many work on cube edge community. Today, I'm going to talk about cube edge, the cloud native edge computing.

A

Platform, this is my self-introduction.

A

I'm a I'm a cube edge maintainer I used to work in k-8s and istio community. I co-authored a book called native service. Mesh is too I'm not focusing on cognitive and edge computing.

A

Today, I talk about edge computing platform built on cloud native, the advantages and challenges. Second, I'll talk about the overview, architecture and core capabilities of cube edge projects concerning how we use cognitive capability on edge computing number, three I'll talk about application scenarios and lastly, given that cube edge, is a open source community. So I'll talk about the introduction, future prospects and the follow-up plans for the community.

A

First of all, here I listed some major challenges faced by edge computing.

A

First, I think the interoperability is poor due to numerous segments of get computing. Second, it's called communication networks, feature low quality and high latency.

A

Often the connection is unstable and the network quality is low and latency is high and often the edge is located on a private network number three edge resources are limited, especially in some campuses or production lines. Small boxes are used, cpu and memory are very limited.

A

Number four. The edge is offline, for example, in campus scenario or in connected car scenario. The edge is often offline, so how to ensure the service is not interrupted when the edge is offline.

A

How do you make sure this application can run smoothly, and this is another challenge that we need to tackle? Next, one would be how to effectively manage the highly scattered edge nodes and reduce o m cost. Next, the last one is about how to leverage heterogeneous resources.

A

Traditionally we have a lot of software and hardware coupled together. Sometimes, if you want to change hardware, you also need to change software so based on all these challenges, when you think about how to leverage cloud native to address these challenges,.

A

These are the core advantages of cloud native.

A

First, cognitive decouple software development for higher flexibility and maintainability.

A

There is a thing called container images. We only need to have one image then, as long as this container is supported, then this can be run in all the places. So this is one advantage.

A

Second, we can avoid it intrusive, customization k8s has good extendability, its apis can be extended, its storage and networks can be extended.

A

So if we make development based on these, it will be very convenient number three. We support. I call native support, multiple clouds to avoid vendor locking for things like k8s. It is uh widely accepted by various cloud vendors, so vendors. They can support this standard apis last one is about improvement of efficiency and resources utilization.

A

It's about using the orchestration to improve the efficiency and resource utilization, and this is kubernetes architecture. I believe you are all very familiar with this. There is a control plane and data plane, control, plane has master scheduler, controller manager and third-party authentication and data plane. It has nodes various nodes and to the right. I've listed some very good design concepts.

A

For example, declarative apis you're all very familiar with this, so I will not go into the details.

A

This is a very good design, which is crd in kds. It can be used to extend apis in kubernetes. You can define a crd and, based on this crd, you can generate development sdk through the controller. The apis can be driven.

A

So you can define a yaml file, you can define a desired state and there is a reconcile logic in the controller which can drive this desired state into a actual state. If you have, for example, you have a three or five instances, then okay kubernetes can do polling until the actual state is achieved.

A

This can be used in device management. We use it to define the iot devices so that we can operate these devices from the cloud.

A

This is another one, very good design from kubernetes. This is a list watch mechanism. If you're familiar with kubernetes you'll be familiar with this one and the left hand side shows the api server connected to edcd right hand. Side is the kas client, so I abstract all the things into client, for example, scheduler, controller, kubernetes or operator.

A

You can abstract them into a client.

A

I've drawn some communication principles once the client is online, then you can obtain the list. For example, you have kubernetes I'm interested in the services and ports running on it. Then I can obtain those from the list then, based on this list, that you can maintain a long connection.

A

For example, you have a port and you update a label, then through a watch event, you can send this information to the a client side. So this is the list of watch mechanism.

A

But in a edge scenario there could be some issues. If the network is disconnected for a long time, then a release, a realist will be triggered.

A

A

Well, send all the information, including redundant information. So if you have a lot of edge nodes, for example, 5000 nodes and 2 000 of them are offline and they go online at the same time. So there is a released operation on the data plane and the system will become unstable.

A

We've done some improvement or refinement concerning this issue. If you are interested in that, you can look into that.

A

This are the key advantages and major challenges concerning the key advantages. These are similar to the ones of a call native, for example, containerized application, encapsulation, build ones, run anywhere. Second, general abstracts definition of applications. This is a de facto standard in the industry. Number three loosely coupled architecture. It is easy to be expanded.

A

The api framework is easy to expand. You use it to define iot devices the right hand, side shows major challenges.

A

First, one limited resources; second, restrict networks concerning private network and limited bandwidth and high latency last one I'll find autonomy needed at ages, edges they never get unstable and maybe offline at any time.

A

So these are the advantages and the disadvantages concerning the platform built on kubernetes.

A

These are some of our thoughts when we design the platform, whether we should use it to manage edge nodes or edge clusters. The left hand side is edge clusters. I have master and node the right hand. Side shows that I have a controller plane in the cloud and I have various nodes at the edge.

A

Actually, we had some discussions with k kubernetes and we conducted a survey. We found that 70 of the users.

A

Prefer to run edge nodes due to lower cost and easier to be maintained.

A

If the edge cluster is run, then the it will consume a lot of resources and it will be harder to be maintained. Therefore, we chose the right hand side one.

A

Therefore, our core architecture is about managing edge nodes, but right now some users. They want to use it to manage edge clusters. So we also had some explorations.

A

This morning we released a new uh open source project called commander. It is a federated cluster and in future, we'll have some joint exploration with that project concerning how to better leverage the edge.

A

Clusters, this is the cooper edge architecture at the top. It is the cloud and at the bottom, is the edge and to the right is the device side in a cloud. There is a standard, kubernetes control plane. We didn't do any change to it.

A

We just need to deploy those controllers and etcd and to the right. There is a cloud call.

A

To address the issue of list of watch, we have a coil in place. It will check the availability of the messages.

A

I would refine the repeated data during released operation later I'll talk more about those components and there is edge core.

A

We did some integration.

A

We integrated multiple modules, so let's just save resources.

A

And in edge scenario, it is easier to deploy edge core. We have a tailored kubernetes called hd hd. It can be used to manage some container applications. We maintained csi and ca and to the right. We packaged some device access capabilities.

A

And there is a mapper.

A

This is what we did to do: device connection or access.

A

Mapper would map the data to mqtt.

A

Recently we developed a sdk for a mapper, so you have your device and your own protocol.

A

There are many protocols in iot. You can have your own protocol and device and based on sdk. It will be very easy for you to develop a mapper in order to connect your device to the system so that you can manage your iot device from the cloud we use a container to deploy mapper.

B

Juicy so this is the architecture of cube edge, so those are key concepts of the architecture of q batch, including five parts. The first is about open ecosystem. We center on kubernetes. We don't change, control, plane of cuba.

B

It means it's 100 compatible with native kubernetes capability, allowing users to use native kubernetes api to manage applications, because many users were concerned about whether it's possible to use kubernetes api. The answer is yes, because it's the same and skta of kubernetes can also be used, and besides we have open list watch apis at edge, which is relevant to kubernetes api, for example. If someone would like to access servers at edge, they can use this api.

B

We optimize this edge api as well, because we have a kubernetes that can be accessed to particles, but now we use cloud core and other parts to do optimization.

B

The third is about the edge device computing protocols like we define of the protocols with cid.

B

Users could expand based on products. We also embed mqtt model bus, bluetooth, wi-fi o p c. U a and z b. You may also define your protocols if you have any. This is to support ecosystem. The next part is about large-scale management of azure devices.

B

First, we have lightweight components, so there are about 70 mb memory for a whole package and kubernetes cooper. Light may occupy 100 mb, so our solution is much lighter and next one is about pluggable edge device management framework and you may deploy the solution with containers.

B

We also support complex edge cloud network environments, for example, when there are some unstable network at x at application layers, we have a message, synchronization mechanisms, it means the message sent to the cloud can be responded by the ad.

B

So it's a kind of to say: I've already received it, and then they were not repetitive, exchange of messages, otherwise introduce kubernetes, and then there could be such a kind of repetitive message exchange.

B

Next, one is about application data at autonomy, let's say in kubernetes cooper light could have their things stored in application, but once network goes this on or you just reboot the system what's inside, the memory will be gone in our solution when there's this connection between applications and edge, and we cannot really permit such loss of data.

B

That's why we have the ash database applications as well as device have their metadata, which will be stored in database and once we disconnect networks and it's time to reconnect again and before the connection and data could be read and the startup applications, so the ming cell edge can run quite a steadily last but not least integrated edge cloud resource scheduling, traffic based collaboration will have very standard kubernetes. That's why we can manage both ad cloud nodes and cloud nodes.

B

For example, we may integrate cube edge as well as kubernetes, as you add a cloud and we may add a traffic at edge and the cloud. Those are the core concepts of kubernetes.

B

Next, I will dive into cubase architecture, which is cloud core. You see the control plan, including several controllers, which are to get the data from cube edge, and then data will be sent to call hack which will encapsulate data and verify it. Today we support I'm socket or quick protocols that will be sent to edge protocols.

B

So this is a cloud-based architecture.

B

So this is architecture for edge. We call it edge call process on the top of it. You see hob, which will receive the message from cloudhub and it will respond to it once it's received and the second one is meta manage, which is to talk with database and then next one is cubelet once receiving data from qlab and the data will be stored in the database um and device trend is to support device connection and bus by mqdd, and the data will be sent to devices.

B

On the left, you see q8s operator, which is new capability. This year we support operator running at x. Someone would like to access api servers, for example csi cni plugins. They have to access edge. That's why we have this operator available which is compatible with api of kubernetes. It means you make access your data. If you can get data locally, you can do that. Otherwise they will ask data from a cloud and get it from a cloud.

B

So now we integrate everything into single one process which is easy to deploy, and it's also quite light.

B

This is about edge device management process. On the left, you see the cids use the fine api through cuber net is sdk. You may have some operations on the right. You see the process graph. Once you set up api, you have to develop a mapper once everything are down you deploy them and then on the cloud you may manage devices at the bottom, so from api to device controller and the message will be sent to the hub and then the devices, the mapper and devices. Through this process we can manage devices at edge.

B

This is edge mass of our community. You may have a question about this because in kubernetes you see a couplexy or cm which can help us to realize services, discovery and traffic forwarding. While we still need to have this edge match, because we still believe at edge, we need to have high integration and lightweight applications.

B

We think it's easy to deploy applications. That's why we have this integration. We provide capabilities similar to kubernetes, like service discovery, traffic forwarding one thing is quite interesting: an edge at each node. We have a lightweight dns server.

B

In this way we can guarantee, once your astronaut leaves the network, we can still have service discovery and the past service, for example, qds may run at the same location in master, but once an hour goes is done and then so the domain cannot be passed. That's why we need to have a dns at each node.

B

So those are capabilities of hatch mesh, but in future, we'll want to add something like install. Api will be used for service governance, which is very popular today in future, will make. We will be compatible with this api. Second, one is about cross domain sub network communication.

B

Let's say in the connect vehicles to vehicles need to talk with each other, but their applications cannot talk with each other directly. That's why we want to offer such a gateway to forward traffic between two vehicles. So this is something we're still exploring today.

B

How to deploy clusters and install it if you want to have a sense of what is cube edge today, we have a key the main available in community. With this, you may install it directly as long as you run. This script run this command. You may put just install it like at ash. What is get talk? It means when you actually try to connect cloud, you need, you need some credentials through get token, and then you may desk something in the command and then there's certain edge.

B

Node will be added, but here I don't have any information about how to install kubernetes, because we're now doing this with standard api, for example, tie the mini cooper can also be possible, and besides huawei's cloud or some other cloud, vendors dividing star, kubernetes and then cube edge can still be added. On top of that, some details about implementation and deployment, for example kdm init through this command. You start this is the latest version 1.6.

B

So all the locks will be stored.

B

This is about edge, as I said, join command. This will command can help us to add this something.

B

I talk about token once deployed something you have to use token to make a required, and then you add that so it means edge cloud will be added automatically and the credential will be assigned to the node and then usually this credential or token can only be valid for one day. It means you can't really continue to can continue use that after one day.

B

Next, I will share with you about what we have implemented in communities. This is a quite typical one. For example, this case uses a cube latch to manage all of surveillance equipment on the world's longest cross sea bridge. It's like 30 kilometers long. There are many cameras equipment deployed on the bridge.

B

There are a lot of weathers like snowy, raining weather, and this platform is used to gather all of the data, and this platform can help us to know what's going on on the bridge, so our community partners here in technology, has adopted our platform. You see, kubernetes and cloud core have been used, so cloudcore is to have management over this edge and besides, we need to deploy those applications and gather data. Through this platform there are some applications developed by the user itself like mapper to support the connection with devices all of the sensor.

B

Data can be fed into that and app qt can help us to make the report and the data will be reported and further to the cloud and so cloud will store the data. The sensing data will be stored on the cloud and then there's also the digital screen, which will monitor everything visualization of the data device, management and other applications of this are things implemented based on our solution.

B

This is another production case, which is quite interesting, so it's based on huawei cloud. This is to manage 100 thousand. As you know, some chinese highway last year and china decided to remove the payment power of the road tow collection stations.

B

Near provinces, whenever you run a car through those towing stations, you'll be charged. So today those astronauts have been deployed at those on those stations, so it means all of those.

B

Trolling parts have been integrated into our platform, and with this one we will know all the applications in our tooling station.

A

For example, you can do one click update or upgrade for all the the etc entries. If there is no such a solution, you need to go into each entry to do the update.

A

So with this centralized management, the operational management efficiency is greatly improved.

A

The etc gantry network connection with the hmrc is very unstable.

A

But with our solution we can achieve very stable application management. The right hand side shows some numbers concerning the efficiency improvement. There is a 10 volt improvement.

A

We can manage these edge applications remotely from the cloud side.

A

In total there there are 100 000 edge nodes with 500 000 edge container applications, for example, charging tooling carplay identification, all kinds of applications.

A

So this is one case study later our community partner will talk about how they implement the solution in their own use case.

A

So after introducing the technologies and case studies, let me talk about the milestones of edge computing in a cloud native, open source community how the community was developed at the very beginning in 2018 in kubernetes community, the iot edge working group was established.

A

This was established by huawei and the open source community and then in november, 2018 huawei open sourced cooper edge.

A

This is the first cloud native open source project and in march, 2019 cooper edge became the first official edge computing project for cncf.

A

In june 2019 we released version 1.0 and in 2020 and onwards had very fast development. Other companies also came up with cloud native edge computing.

A

In september 2020, we.

A

Promoted our projects from sandbox project to a more incubating project.

A

As I said earlier in september 2020, our project was promoted to an incubating hosted project of cncf.

A

There are many developers and contributors in this community and we also have a lot of contributing organizations. The right hand side shows the ecosystem. We have sick.

A

Focusing on three directions: one is a device iot. This is used for device management. Second, sig, ai.

A

This is for hai.

A

For example, when it comes to intelligent driving, hai is very important and we have a sub project.

A

This is a used to address the edge cloud synergy, ai issues next speaker we'll talk about how cetena was used to address the hai issues. Number three cmec: this is for 5g access, so these are the three sigs. If you are interested in them, you can follow us.

A

We have a regular community meetings.

A

You're welcome to join our regular meetings.

A

Community version release pace, we made a major release every three months later version is 1.6 and in three months we'll have 1.7 between 1.6 and 1.7.

A

We would release some minor versions to fix bugs and we support both x86 and arm, because at the edge there are both arm servers and x86 servers.

A

So we support both of them.

A

These are the community partners and users. I have not list all of them, as you can see from the slide. There are so many partners and users.

A

The right hand side shows various application scenarios.

A

Future prospects.

A

Concerning infrastructure.

A

In future, we will make some exploration on the edge clusters concerning platform capability.

A

uh Edge cloud traffic cooperation and maintainability, observability and serverless design: this will be our focus. The fields.

A

And business domain for things like internet iot devices and a I ai edge cloud synergy.

A

A

Primarily on access, lastly, concerning community operation,.

A

We're now running a an incubating project in cncf.

A

We welcome all of you to join our community, we'll add new maintainers maintainers will be involved in code check, code review and code checking and we'll have joint promotion with partners and community case sharing concerning how people are using those projects and implementing those projects.

A

So that's all for me today. Let me do a recap concerning cube edge.

A

This is a cncf incubating project.

A

We have open community governance which allows the users to participate in community development and governance. For example, the version planning and development of versions are all planned by the community.

A

Participants and we support the dynamic device access. We are compatible with edge cloud synergy. We extend the cloud capabilities, such as ai collaboration, data collaboration, application collaboration and management collaboration to the edge and cloud edge. Communications is about high speed, reliable and secure communications between cloud and edge, and easy maintenance is achieved by lightweight and plugging based edge framework and, lastly, we support heterogeneous hardware. We support all kinds of hardware, there's a decoupling from the hardware so that we can support all kinds of hardwares in the scenario of edge computing.

A

That's all from me today.

A

This is our github address.

A

This is the slack address. There are many developers in that community.

A

We have the huawei maintained public accounts. You can scan the qr code to follow our public accounts concerning the container magic cube, and you can also scan the code to the right hand side to join the group. The wechat group.

A

Lastly, this is our training camp.

A

Concerning the training on cloud native, you can scan this code to follow our training camp. We have a lot of experts delivering the trainings.

A

That's all thank you.

A

You can scan the code to appraise the course.