 Good evening, everyone. I would like to thank you very much for being so patient with us to now. So first of all, self-introduction. I'm Jiaxuan from Intel System Software Department before original name is ODC, working for Acron. So several questions. How many of you know Acron? Raise your hand please, a lot. Thank you. Second question, how many work for industry, sector, industry? A few, only one. The third question, how many working for automotive or onboard vehicle? So I need to share with you a lot of information, otherwise lots of information would be lost because these are relating to the three questions. I won't have 35 minutes, so I will deliver my presentation within 30 minutes. This is the type of contents. First is Acron, what is Acron? 60% of the audience today, no Acron, 30% do not know. So I will introduce this briefly. Second is industrial use case. Last year, in other places, North America, Japan, we talked a lot about onboard vehicle. So that was a focus last year. And this year we will focus on industrial example cases. But of course we will continue to work on automotive. And industrial would be our expanded area this year. And the part three is about architecture. Part four, roadmap. Our roadmap of technology. Part five will have a link, including WeChat group. So I will provide a QR code and you can scan to join us for further connections. Also I have lots of extra information for you to have deeper dive. The details of technology is depending on how much extra time it will have. So first of all, what is Acron? Pre-start, sorry. Sorry, technical issue. What's Acron? It's a flexible open source lightweight type of visor for IoT workload consolidation. Also in 2018, Linux Foundation will launch this project until now. It has been carried out for about one year. Before open source, our team developed this for about one year. So from beginning until now, it's about two years lifetime. During development, we set lots of targets. First of all, servers and technologies. We need to set this apart. Otherwise, we do not need to create the Acron. And several differences. First of all, small footprint relating to size. The current code is about 31,000 lines. If you have a look at KPM, it's about millions or even tens of millions lines. So if this is for commercial enterprises, the scale would be far more than 30,000. Second, we mainly focus on IoT. So lots of special needs, including what I mentioned before, footprint. The second is within IoT area. This is heterogeneous workloads consolidation, including real-time and non-real-time. Functionality-safe and non-safe. So two different types of business to be consolidated. The third is our license is BSD license, very flexible. And different to the others. We also have the other key values, including system security. So we have a VT to back the virtualization. So we can pack Acron. And to schedule this, also we can use Kubernetes for scheduling and orchestration. No problem. We have our internal codes and not all open source now. So these are the positioning for our value. This is about our progress. Last year, we mainly focused on automotive. The main scenarios are console to be integrated with the infotainment. So standardized two platforms. One is console, another is onboard infotainment. Lots of customers also use three screens or four screens. We can also support this. So in May this year, we launched V1.0 version. We named this as Acron 1.0. This is also a very large milestone. On the basis, our clients have introduced this into the product. For confidentiality, I cannot name our clients for you. But in the future, you will notice our Acron technologies in the vehicle. No more details to share with you. Also Android and the other OS, we can also provide support. I will focus on this topic today in terms of industrial side. Let's have a look at the scenario. For industrial, there are lots of uniqueness. Let's discuss the scenarios first. Just now only one of the audience today raised his hand and is working for industrial side. Currently, the consolidated parts are a few. One is real-time applications including PLC, CNC, robotics, and real-time applications like this. Another part, workload interaction. In the next line, we have a console for the workers to operate. The machines can be controlled. For example, the numeric controlling, the tooling can be controlled by this console. In the conventional scenarios, we have two independent systems for operation. You can do this on to the same tooling. If you participate in the industrial trick shows, you will find that this is a small console and have a black box to put it in the front. Also, it will have the windows. Not flashy, not very appealing, just very simple programs. This is X, Y, Z coordinate and the program operation situation. On the back end, they will put the tooling. So this is very typical scenario. Lots of manufacturers work for the councils. They need to purchase the products from Japan, from Mitsubishi, or from the other overseas companies. The cost is very high, very expensive. Looking to the future, can we consolidate these two parts together? Which is real-time applications and console. Industry side, we name this as the consolidation among the two. Currently, we don't have multiple solutions to choose. I know Siemens has their solution and another manufacturer. So in this area, they have very fierce competition. Who can win this market, their products can be improved a lot. Today, we mainly focus on the consolidation and preparation. You can see here, normally within the chip, how to consolidate these scenarios. Normally, we do like this. The typical scenario is like this. As I mentioned before, the council and the windows OS. Windows in the operation condition covers about 60 to 70% of even higher market share. In control, we use VIX works for high-end. In China, we also have lots of domestic clients. They use Linux-based OS to reduce R&D costs. Because VIX works, price is very high, and license fee is very high. In addition, another also works very well. Very important safety. IEC in the industrial is very important certification. It's very difficult to be certified by IEC. You have to guarantee the safety level, at least in the production line. To avoid fires or downtimes or the loss of expensive components. So safety is very important. In terms of automotive sector, ISO 26262 is a very important certification. So we will focus on these two sectors today. This is the design of our architecture. And the things to consider normally we do not like this. First of all, the low resources would be segmented, would be categorized into two types. We have ZFI. Today we also have ZFI topic presentation today before. We use it for safety guarantee. It can be used to monitor the hardware condition, to check the healthy condition. If anything happens, it can send alerts in real-time. So safety alert. On the right, the partition. So this has two partitions. It at least supports one tools. And it emulates multiple PLCs or CNCs, whatever. HMI normally only has one. Depending on CPU capability or the quantity of resources for HMI. We will reduce the resources it consumes. So it will go through service. We also provide sharing services. And the back end will support this. What are the difficulties we will come across? First of all, mixed criticality. So it integrates real-time, non-real-time. And giving us lots of complexities. Because for real-time, this needs to prevent for the interruptions. If you are executing the events and accessing cache. Windows also wasting cache, accessing cache. Our three cache, then the cache will be influenced. Maybe within the cache, there will be influence. And for next-time access or execution, this data or commands needs to have some changes. And the real-time will be influenced. So this is the interference. The biggest challenge for us to deal with, the second is safety. So we regard this as safety domain. So we have passed this certification for safety domain. We cooperate with most of the certification agencies. And for the right part, very pricey, very difficult to pass certification. We have accumulated the experiences. Normally a line of code. The certification cost is very high, 100 to 150 US dollars. So it is not possible to certify that. And the system will be divided into two parts. The safe and unsafe parts. And here it might be interfered. The unsafe part may affect the safe part. How to address this issue is also important. And then about isolation and sharing. Of course, we want to share the resources. Otherwise, why do we start workload consolidation? On the other hand, we want to have isolation because we don't want other OS to affect our tools. Otherwise, the components may be problematic. So the best way to avoid this influence is isolation. You have your own storage and your own network. And this is the best way. But in our system, it should be implemented on the same level. And of course, maybe some interference will be generated. And we also have some other decalters. Like the runtime, the runner-trick time. And also the MSI interact latency. And also the Xenome latency, which is always about to measure its scheduled latency. And usually for the task scheduling, the density will be around one millisecond or 10 milliseconds. And then the HMI is Windows 10. We have designed this architecture on the left for safety OS, which is pre-launched. It means all the safety OS. It is called the pre-launched safety OS. On the right, it has a post-launched. And the service VM in the middle means it starts before that. And on the right, it means it starts after. And for the post-launched VM, it is referring to the service VM, the Akron device model to start it. This is a basic concept. And for the four models, how would we use them? For the one on the left of the CF safety OS, it could be restricted by the watch machine. If that stopped working, the safety OS on the left could still continue working. Like the operators could then intervene. And on the right, the RTVM, real-time VM, to run the real-time OS, is to run the control codes cycle by cycle. And then to operate the production devices in the factory. And then about the devices we support, you can see many of them in the list, but some highlights I would like to point out. TSN will be used in the industrial scenario. Eastergate and the profanite would be used more often. Like for Siemens and Profan, Japan would use this. And these would be based on the private protocols. And for TSN, it is a standardized protocol. Like the private ones, it is a public and unified standard in the industry. And its advantage is that it could help achieve connectivity between the products. And for TSN, it is to control the device. Like the TLC would throw TSN to send the network package. And then to drive the motor. So the network must be real-time. And the TSN is a time-sensitive network. It's for time-sensitive network. And with virtualization, you could achieve the real-time effect. And then about FPGA, it might be needed for some mathematical processing in the industry. And usually for FPGA, we would have it cut through through RTBM. And then directly access FPGA. And we also have some customers asking whether they could be shared. So that they would support multiple simulated PLC to visit the same FPGA. But now we do not have clear answer to that. If you want to virtualize FPGA, it's not that difficult. But if you want to make it real-time, it is quite tricky. So we may need further design on FPGA. And the next one is about GPU. For the onboard infotainment system, they need to share the same GPU. But in the industry HMI, it's not the case. So GPU is mainly for Windows. But it is not for mediated path through for the GPU. The real backend is in our room service, device service. It is for cheap HMI and Windows. The main reason to use this is to avoid interference. If we just directly use it for Windows, Windows is not that reliable. And some people may think Windows is a malicious system. And it would lead to some problem. And if problems happen, it may cause collapse of the whole system. And the graphics driver under Windows, then the graphics may be collapsed. But the probability is quite low. But once it happens, it will lead to the failure of the whole system. And we want to address this issue. And other devices, we'll just skip them. Let's now look at the safety problem. In the industry, we need to meet the IEC 61508 standard. And here we divide the system into two parts. On the left, the below part is a hypervisor. Here we consider about Zeifa. And here we use a blue color to frame it. It is certified by Intel or customer. But Intel will not finish everything. Some details will be left to the customer because Intel is not offering end-to-end solutions. Customers need to cover part of the job. And they need to certify it by themselves as well. This is the left part, which we call the safety-related part. And on the right, the purple, the pink part is not the non-safety part. They may run the Linux system or Atooth. And this is not about the safety certification. And we want to make sure the right part will not interfere with the left one. We may need to try to avoid the right part, the non-safety VM codes to prevent the non-safety codes to affect the safety ones. And I will not dwell on the details because the other day we had a meet-up event. When more than 100 developers attended so we also published our tutorials at that event so that you could know more details. And then for Windows as HMI domain, in the industry, Windows is quite dominating. And maybe people will ask how about Linux. Linux is easy to operate, but the main thing is in the industry for the ITers. If we want to track a problem that's easy, but in the industry, it's not the case. Maybe they are not the ITers, they are like mechanics, and they don't know about programming. So maybe those codes used to run on Windows 7 or even earlier editions, and the compatibility may be not that ideal, but reliability here is quite critical. So we want to make sure that with the upgraded system, the codes could still run. The compatibility is quite important here. We want to make it here mandatory. And if we want to make Windows, we need to first of all, for starting we need to adopt OVMF. We want to avoid change to Windows because it's closed source. We have the OVMF, we copy driver, and the logo could be normally demonstrated during starting. In addition, we also support a secure boot with TPM, and we will adopt a watch TPM, the virtual TPM to do it. So for the secure boot, it is like this. And for Microsoft, for virtualization, they've initiated their own framework, which is called Hyper-V. We will not support the whole set of Hyper-V, just a subset, which is called TLFS. The Hyper-V, Hypervisor Top Level Functional Specification, we would support this one. And another example would be, as I mentioned, Windows is a malicious OS, and on our Intel hardware platform, we also have GT-class, a GT-class for the graphics. It could access to part of the cache, and it may interfere with the OS. And Vixworks and RT-Linux to realize real-time work of the key technologies. First, we would path through the local key to the RT-VM, and the interruption of the processing is just to interfere with the OS, and the Hypervisor will not be intervened to reduce the latency. And the second key technologies enable cache. The main feature here is to divide the cache, and on the RT-M, it will have exclusively occupied cache, and will not interfere with other OS. And in our BIOS, we could turn off some feature that will influence the real-time performance, so as to improve the real-time performance. But we need to have the customized BIOS, and we need to follow a guideline to customize it. And we need to offer these options so that before shipment, the option would also be closed to avoid further unnecessary influence. But how about its real-time performance? We have evaluated the real-time latency. We have the Intel Core CPU with 1.9 GHz and 8 GB memory, and we also have the premarty for both, and we would benchmark both on their configuration, and this is the result we got. On the left, you compare the upper and lower graph, and you can see how it is distributed, and the closer to the left, the better the jeta, and on the right, it is the overall jeta, and for the jailhouse, you can see it is closer to the upper part, and the lower one, you can see at least about the achrom data, and in our 2019 results, we have some Intel policies to maintain them, and this is also derived from the achrom one, and we also have our jeta paper published for your reference. This is our roadmap. You can see that both green and red parts are about our real-time requirements, and by the end of the year, we want to activate all the features. These are our major goals, and we are approaching the end, and here you can scan the QR code to join our WeChat group. Actually, this is my personal WeChat account, because our group is big enough, you cannot join, and you can just scan the QR code at me on WeChat, and I can introduce you to the group. Any question? Interpreter, can't hear anything? It's all on expenses combined for development, for construction. I didn't quite get you on the safety issue, so are there different levels for the safety? It's cell 2 now. For cell 3, we need more redundancy on the hardware, so it is not isolation physically, and what is your fundamental method for the certification? Well, let's go back to this slide. If you go to see the ISO standard, it establishes what conditions you need to meet, but here for our certification, we will try our best to... Put safety OS on the left to run some hardware codes, and we also have STL-TOM hardware to have regular examinations, every 125 ms. And if problems happen, it will give a warning, but how would you guarantee the time sequence and storage of the directives will not influence that? Well, this is a very good question. This is what we are making efforts on for the watch machine, each and every directive, and also the temporary isolation, all should be analyzed, all codes should be thoroughly analyzed to make sure there's no problem. Every three lines of codes, it needs a safety requirement. I think there are so many scenarios. Yes, and two-thirds of our human resources are put on this. It is not $100, it is between $150 and $200. Maybe you can have two pathways, and then you decide which one to choose. This is our next step of action, to realize cell 3. And if we put the codes into this boot load, we would have the QA of the storage, or we need to make sure the hardware status should be tested to make sure it meets our expected condition. So the bottom layer, at the BIOS, made a lot of settings. Thank you, bye. Why can't I open my mouth?