Red Hat OpenShift Telco | OpenShift Commons, 29 May 2019

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Case Study: OpenShift and Edge Computing at China Mobile - Xian Jia (China Mobile)

Description

Case Study: OpenShift at China Mobile
OpenShift Commons Gathering
Barcelona Spain 2019
Speaker: Xian Jia, Project Manager, Edge Cloud Computing Architect (China Mobile)
Title: OpenShift and Edge Computing at China Mobile

A

A

I'm I'm from China Mobile, my name is the chuan is very pronounced. Hardy I'm, the edge cloud computing architecture for this presentation. I will to tell the story about the edge computing with OpenShift. So so I want to tell the background. So what is H computing? The is computing is a computing and network deep fusion. So I draw a 5, a 4 pictures here, so we can see which is the H computing. So we can check the red star.

A

It's for the Aged, the environment. A is very little. Its didn't like the data center, which is, can concern 1000 per machine in the H in H H computing center. It can only come only have a 10 or 20 parameter machines, so the resource is limited and I think. This is why we want to use a darker, a container technologies in this field, because in the telecom field we almost used OpenStack, but now I think this is the community's opportunities.

A

So in the age a we celebrate, we Sturbridge 2 part when the networks network side, each computing, which is below the awareness station and for the other side, is computing instant. Like the kid way we can put it in in the home or in the industry, so for the in the age computing we can provide the the eyes the past and the sass in telecom field. We are always followed the ETSI or some others international standards, but and now I poured into this the this architecture from ETS that we see. So it's very calm.

A

We can see in this picture in the left hand, so there's a lot of the interfaces in here and they are not defined clearly now, but in China Fuji and the cloud knee under it. Computing are very popular, a lot of company like the Baidu Ali, wahwee, DT and others vendors.

A

We want to use these technologies in China and in this in his right eye, a laser some, the pros and cons so for the pros it is based on the NIV architecture and it have defined the network infrastructure in information and the location, API, new identity, API and bandwidth management. So maybe all the people in here I don't know what Hayes is so I can give us an example. So when we open our phone, so you will see there is a 4G at the at the top of the our screen.

A

So all these only a movie clip at the Chrome application. So we can search the e tonight. So already in title flow is in the telecom cloud. We call it it's some like the private cloud. The oldest data flow is the enterprise private cloud, but in the age computing, who is our customers in China? Maybe the by to aunt, isn't Alibaba and some industry company.

A

So because all these vendors are from a node in the telecom field, they are our come near ours customers. So we have to provide a ties. The cloud public cloud, but in ETS diamine see a standard. It sound like profits.

A

It sound like a today defined like the private cloud, so it's not satisfy our requirement so I gave the conclusion here is the I mean the average and IV architecture is to complicate, and there is a far away from our product, but we can use. The API which is have defined in the idiom is e, so in the LAV field and we choosing in the telecom companies it's all we all we all used OpenStack, but in its computing we want to use coupon IDs. So I try to refactor the architecture from the ETS imec.

A

So we can. We start we split to field when IV field. This is the telecoms things, so we we don't care about that and another one is the EC field. Is the field at each computing field?

A

There is the H computing operation managed center is some like the photo and can manage a lot of the did each computing a cluster and again provides the beam the container and the stamp third party API unfold nmv. It only put fears atrophic into a in into the age computing data center. So we so there's two part is ace.

A

Is a beaut mmm it's billed separately, so why we want choose open shift? I have lists to the four regions here. The overshift is on the most major of installed project with support communities nearly before I joining the China Mobile I'm I'm. Worse I worked in your head of four years ago, so Amber's a red Hatter, so I know it is the very stable and now we start to develop a project based on the kubernetes. So we choose the overshift as our basically kubernetes an oval-shaped and provides at least ten layers of the continuous security.

A

The container a host and multi-tenant container platform, network isolation and so on. An open ecosystem is a flourishing, so I know there is a lot of the telecom and the band companies and industry companies they are the loads coming in are using so they have lower.

A

So they can so they can use use a few effort to move their application into our platform, so we can build up. Who can build up our own? It computing data center and open our urban ships always brings amazing features from open source communities, so seems a lot of the companies have using the urban shift of four four or four pound. Oh, we are, but now we are using the communities three pumps at the other, so we have to catch up with other guys so Sigma 8h computing platform in China bebel.

A

So we expert, we have low-dose experimental environment in China, for example, in Georgia and condom and beating.

A

So I, so we can see this. This is our attractors in the red part. Is our ace edge computing operational management?

A

She may be it. We can think it is in the in the center of the area, for example in abating, and they can manage a lot of the problems in China and for the green part, is a is our age I'm huge in the center, so it can run on tin, beeramid, machine or 2020 pyramid machines, so we developed some additional modules, for example, is the classroom management and service routine model?

A

So it's not like the openshift routine, because it's not satisfying our requirement, so we developed it by ourselves and undoing here's, the third pratik capabilities, industry, sdk awareness and network capabilities. This is our from the industry, so it's very company, it's very complex. We do it back his keys back ease and we are using. We are trying to use three skill to manage our api, because in some industry we have to provide api and then we can, we can serve the api and and and on the money you can give the earth the billing unfolded.

A

I read for the yellow part, for the yellow part is a the edge computing gateway, so it was smallpox. We didn't use. Openshift I will use the cout page because very small, it's very small, so standing like the the little box. Maybe it's wrong in the cell phone, so we so open shapes is 3.0.

A

We use the lot of the GPU capabilities so now I want to try I want to introduce the use case. We do we do we collaborate with the by two companies- the use case, auto vehicle in this computing.

A

So we talked with the Baidu company who has joined their own Otto Michaels in Beijing, and they provide to the requirement and the we build. The arch actors. So at top of this is a beta cloud. It sounded obvious, but I do has their has their own public cloud and there is a Sigma Sigma is I is just totally at Otis to talk about it. It's our age, computing platform and it's the bottom and at the bottom, this this program we can see there is a 5g network and some the aware network.

A

So the procedure, the workflow is a video stream is saved in the H node and Sigma, and the AI, which is the application running in the Sigma and the AI, operates the video stream in each node and then sent to the read out to the to the ICU I see we can record, we can go away a regard. We can think a T. Is the machine in the inner car understand the uncertain pack, the data into the PI to cloud?

A

This is public cloud and I see you I still send it out to the car as the PC vibe interface. So this is the the architecture we test weight has in baking in Beijing, so one we choose two ways: one is the Fuji real news network. Another one is the wire net. A wire network is some like the way provides the VPN to two by two, because we want to to to balance if the Fuji, the price, the price is.

A

If using the apology, the price is too high, so they can transfer it into use the wire network. So we balance the two technologies about when we test it. We find that we found that the where net it's too slow, too slow and it costs a lot of time.

A

So experimental environment here is and provided is application because obey to the applications always continue eyes and the I. The applications have already run running in the communities, so we have whistle it can easily run on mobile shift.

A

So the requirement is the banner. Is there? Is a 6th Mbps, a blue stream for camera latency, so why we want to use the edge computing because it can provide low latency. The latency of the data flow from camera to age computing server is the less than 100 millisecond the latency, with edge computing a server give the result. The Houston to the response is the less than 50 a network. Latency is less than 2 miles purely understood to disinterest.

A

Here, when is the edge computing the center and another one is a public cloud cloud to the center.

A

So the reason we worry complex so Hanks into that deterred to to the car and since the key data to the public cloud and and in the future they are gelatos of 8 computing in a center. So all this application are distributed.

A

So in this experiment, environment we choose to Street Internet intersection and employ to totally 26 camera there's a lot. They are interesting in Barcelona, I didn't see any camera in the street, but in China just lots of the cameras, so 26 camera in to Street intersection, and it calls a lot all the data.

A

The Cameron Nico needs the electron-electron the power because when because they had our the the machine, teenies need their power, but when we, when we set up the camera in the light, London single light, who provides the the leg electronic power so how to talk to with the garment and the government, provides the the Leghorn Ingo power. But it will cost a lot of money.

A

So the schedule is here: we in February infra in in fabric. We.

A

Is a d2t design, and now we start in testing and debugging in still in the testing under testing now and in the next month. We will finish the testing.

A

Here's some pictures, you can see the cameras, there's a single light, so I want to introduce some challenges when we are working on it. So there are two challenges here: Alice today, when the storage, wind, the GPU, so just a mention that each camera will upload 6 Mbps.

A

So there will be 64 GB de disk, only one camera, but we have the 26 cameras, so it will use modern 1pp per day, but but by do calmly told that we have to keep three months the data it will cost 100 TB, but I just mention that in the H computing data center there is a few very much machine here, but so this is so huge storage.

A

So we have to use some other technologies to to reduce the data in each computing, so why we choose to send a video stream instead to image. So we we talked with bed companies and they said that the if we use the video we have, we can use some some model modern, some new technologies to truly to to reduce the the data floor, but but yeah, but one second, a most 1mb.

A

So it's so lot for lunch ham, so we will sink it later and the talk wizard about to if they can use the I use image instead of another, is a GPU in because in the age computing the data center, the the machine is not like, as not like in at the machine in the in the data center. So it's it's more, it's more.

A

We call it the oti, which has with find ways channel mobile and other companies, something like like the by to Ali Baba and the because of the the machine is, is more a smaller, but for this Expo is experimental.

A

We have we have. We have ensured that each server needs at least three GPUs, so stretchy peels is okay, so the machine can have the three PCIe. So you can put the three GPU, but the GPU cost a lot of electron in power, so we will need to reboot our eight computing center, because the edge computing Center is not like the the the the computing Center, because that, because the edge computers in term maybe is near our building, our home near the our industry and the electronic is limited.

A

For example, if in China there is a old building and can only support three key W I, don't know how to pronounce it. Okay, three key W, but if we put a lot of GPU they did the datacenter let each the center can support it. So we have to reduce the resource we have used in H computing, but in the to the center we don't match these problems. We have lot of space, it 2 2, 2 2.

A

We have a lot of space to put our pyramid machine in here and they were doing to worry about the electronic power, but in it computing we have choose thinking things like how to reduce resource the container technology is a better one. So what we want to use it open shift here and to be continuous, that we are doing a lot of the things in the future. So now we collaborate collaborative ways that tritium companies. They were two prior, the API and the China state construction.

A

They provides the EPP the at the application, but we, our application, is running on the Virtual Machine.

A

We only have three pyramids machine, but there we built the upper shift here and enabled Robert to provide the wash machine and give the the virtual machine environment, so they can put their application. Monument and other is ID by two companies and the blockchain and the in the next month will deployed all the things in Judea. It is a 4G amidst the environment and in the the later we will he was the overshift boy.

A

Oh yeah seems we are late simulate other in in the August we integrated with the acronym, which is a project in the linen Phoenician age.

A

So as computing is a very popular in in China, but teams in Europe and the United States consume is it's not seemly? It's not, but Foudy is so popular. So we think the fuji and it's computing will bring more valuable things in our industry. Sex for.