 Good afternoon, everyone. Welcome to attend the session about the Cyborg project update. I'm Jinghua, and my English name is Coco. And I'm here in the representative of the former PTL, Howard Huang, and the new PTL, Li Liu, since Lee can't make it to Berlin. And here's the agenda. Since Cyborg was a very new project, I will first introduce this background. And then we will take a look at Cyborg's journey and its achievements in the past two years. Next, I will focus on the rocket release features and the plane in state release. And the last part is just some ideas which elaborate why Cyborg will continue to be of value into the future. And about the history of Cyborg, we start from the technical requirements. But later, we found there a bigger scope, like in public cloud and HPC. And we first named this project, NoMend, from the upstream discussion, and then changed the name into Cyborg and a startup implementation. Another aspect, why we need Cyborg may be the common errors of the domain-specific architecture. Since more law can hardly be followed in the future to achieve higher efficiency and performance on computer storage and network areas, domain-specific hardware will be highly demanded. And examples are like GPU, NPU, FPGA. And NPU stands for the neural network processors, which is used in machine learning. But DSA is facing the fragmentation problems. It's really hard to manage it due to the difference in its architecture. And Cyborg is here to help. And that's why we started our journey on acceleration management. And it was Howard Huang at first 10 months who started this project. And then Russia and Justin joined in the next eight months as they built their initial architecture for Cyborg project. And in the last one year, the numbers of the contributors have been growing so fast, you can just get the number. Yes, 36, the contributors. It's really hard to believe, right? And we think the number will keep growing. And these contributors are from many different companies, like Huawei Intel, Lenovo, Nynan Cloud, and so on. And there are many joint efforts which came from many other companies like Invadier and iFlight Tech. They really gave us advice on how Cyborg can be used in dedicated usage scenarios. Indedicated, I'm sorry. Also, they are growing maturity of our project. By this September, we have nearly 20,000 of IOCs. And these efforts from the upstream and industry all drove the evolution of the Cyborg architecture. Right now, we have the DB Conductor API, which is placed on the controller node. And agents and different types of vendor drivers are placed in compute node. Also, we support FPGA programming and the resources quota. Besides, in order to interact with NOVA and other projects, we also add OSACC library and the Python client. And that's what you saw in the right. It's our newness architecture. And in the Rocket release, we apply access and quota control over resources. And for the driver, we work with various vendors to standardize APIs for the heterogeneous resources. And besides, what's worth to mention is that you can use our FPGA APIs to program the Intel FPGA. And for the interactions, we provide the Python client for the users and other projects. And also, we discussed with NOVA team on scheduling procedures of VMs after using the Cyborg. And in the state release, most important thing we are going to do, I think, is to make Cyborg functional with VM boots. We have discussed with NOVA team in demo PDG and make an agreement on how Cyborg interacts with NOVA. To do this, we first add new concepts and features, which is called ARQs and device profile. The ARQ stands for accelerator record. An ARQ object is used to encapsulate the assignment requirements of an accelerator to an instance. And the device profile is used to specify users' needs for different kinds of accelerators. To support a VM with required accelerators, users must specify the device profile and the NOVA flavor at the same time. And besides, we need to add OSACC library. The OSACC library is similar to OSBreak, which provides an abstraction way to attach accelerators. And we need also to implement the DB schema for the ARQs and the device profile as well. And for the drivers, we are going to expand different vendor drivers, like Xaninx, FPGA, Huawei, NPU, and RISC vDriver. And if time is enough, we will try to propose ML2 plugin, probably named networking Cyborg, and add notification support in Cyborg. By doing this, SmartNIC may be managed via Cyborg and used by Neutron. Apart from the open stack integration, they plan to integrate Kubernetes as well. And to do this, we need to first align our data model with the device plug-in interface. And through the DPI, which is based on Kubernetes master, through the DPI plug-in, the Kubernetes master can talk to Cyborg, so as to manage accelerators. So I've been talking about what Cyborg done at, what Cyborg project done in the past. And so what about the future? Why we keep working and put on our enthusiasm on Cyborg project? I think that maybe they believe in some of the future proof ideas. And the first one is that they believe that AI were got involved maybe in most of the life areas. And they really did a full-stake hardware and software on the hardware. And from the bottom to the top, they are DSA hardware, tool chain application framework, and high-level applications. And TensorFlow, as you know, is in the application from framework layer. And we believe that in the tool chain layer, that is where Cyborg can help. And Cyborg can help to do the resource management or scheduling of the different DSA hardware. And the second one is that acceleration-centric computing, which is another view of our data center computer. And the Microsoft proposes their understanding of a new data center computer. And as you can see in the center, in the middle of this framework is FPGA rather than CPU, the FPGA is responsible for directing workloads to all their traffic. And we believe that not only FPGA, but also the general heterogeneous hardware can be used in this center. That will bring another needs for Cyborg. Yes, I finished. Thank you all. So if anyone has questions about Cyborg, am I talking too fast? Oh, please. Yeah. In Rocket, we haven't finished the integration with Nova. Yes, but we are trying to get that done in the state release. Yes. And therefore, if you want to know more details about how to interact with Nova, we have specs in our website as well as in, I think, the Nova website. Yeah. Any other questions? Oh, thank you. Yeah. You know, you mean this slide? Right now in Neutron, it can manage the SMANNIC VR, like the SRV agent. But the SMANNICs, they do find it have some problems. Like if you use it, you have to record what VFs you have and what others may not use it, you have to record this. And we find that the Cyborg can do this new function. It was just they proposed this like integration playing in the demo PTG. And it was just maybe not that mature at this moment. So I don't think. Yeah, we do have spec. We have spec, but that wasn't done at this moment. I think there will be a road in state release. And I think if you are much interested in these questions, you can refer to Thundar and his work on it. His work on the proposal of how Cyborg interacts with Neutron. Anyone else? OK. Thank you. Thank you for a 10-year session.