Facing microservice innovation, IPU is becoming a dance elves in the future data center

In the mythology of the Germans, there is a kind of elf named Gremlin. It is said that this kind of elves are familiar with various tools and machinery, which can help the skillful craftsmen to complete various work, and can improve the technology of craftsmen, inspiring their invention spirit. Many times, the traditional methods and methods of craftsmen reach a certain bottleneck, and they will pray that the elf appears and can help them solve the problems.

In our era, the data center is a great craftsman, and shoulders the heavy responsibility of creating computing and intelligent infrastructure for Qianxingbaiye. When the data gravity encounters the work bottleneck, the elves may also need to join their work.

Recently, Intel joined hands with Baidu Intelligent Cloud to release the "Provide an Accelerated Engine for Infrastructure to help shape the Future Data Center" Intel IPU platform solution White Paper, which mainly analyzes the functions and applications of Intel's IPU architecture, and how to cooperate with Baidu Smart Yuntongli to create Its new DPU architecture helps cloud data centers to fully release computing power and comprehensively empower cloud service innovation.

In the era of AI and big data, the cloud data center has become a top priority. The release and practice of Intel's IPU can maintain the high performance and high utilization rate of infrastructure, while achieving more flexible resource allocation, and enable cloud service providers to turn to complete virtualized storage and network architecture faster, thereby greatly satisfying great satisfaction Cloud data center upgrade needs.

In the White Paper, we can use the development and challenges of the cloud computing center-IPU architecture and technical advantage analysis-Intel and Baidu Smart Cloud's practice, three dimensions to examine the development blueprint and value expectations of IPUs.

In the data center, the elf is waiting to dance.

Cloud Digital Intelligence Reform: Pressure and Bottleneck of Data Center

The cloud era is booming, and new technology and industrial changes are brewing in the cloud computing industry. For cloud service providers, corporate innovation applications have the ability to put forward higher demand for cloud data centers. With cutting -edge technologies such as 5G, AI, and edge computing have obtained landing applications in more industries, many new changes and trends have emerged in enterprise -level cloud services. For example--

More distributed cloud services deployment: The development of technologies such as edge computing promotes cloud services from traditional central clouds to gradually evolve to the "cloud edge end", and more cloud services are deployed on the edge end of the application.

More extensive cloud service application scenarios: cloud services are becoming one of the IT infrastructure of more industries. It is the main force in new formats such as intelligent manufacturing, smart retail, and cloud games to support the digital and intelligent innovation of enterprises.

In the promotion of these changes, the cloud data center server is facing more challenges in management and applications.

For example, the deployment of more virtual machines (VM) makes the management tasks of virtual machine management more complicated. At the same time, its capacity output has gradually developed from traditional single applications to microservice, which has further improved the complexity of cloud data center management tasks, thereby occupying a large number of processor resources.

It is necessary to carry out efficient management of virtual machines and microservices, but also accelerate the realization of infrastructure such as the network and storage of related cloud data centers. The complexity and resource overhead levels have continued to increase. Data center requirements.

As micro -service models are increasingly used in the application, storage and network workloads of cloud data centers. What can predict is,

In the future, data centers need to be transformed and upgraded in the following aspects:

1. The emergence of cloud native applications and rapid growth have pushed up the demand for special infrastructure, and greatly increase the agility of cloud services.

Sex and efficiency of cloud data centers.

2. The microservices of cloud services have promoted the development of a distributed heterogeneous computing environment, and each microservices should be more suitable for it

Run on the acceleration node.

3. The extensive adoption of microservice models has also spawned the data center arrangement system to enable the automation and management of microservices to distribute microservices between different structure computing servers.

4. The use of microservices, virtual machines, containers, and containers has continued to increase, which has promoted the development of service grids. The service grid can simplify the communication from microservices to microservices and make it more efficient. The service grid has become a standard component of the Yunyuan stack.

Therefore, the architecture of the data center in the future will be designed with more grounds to microservices and increasingly decentralized application environments to better use the accelerated processing capacity brought by heterogeneous computing. Overall, the future data center architecture needs to be designed for micro -service and decentralized application environment. Through heterogeneous computing methods, the data center computing power facilities have their duties and adapt to the characteristics and needs of distributed deployment. The industrial upgrade pressure of the times.

In this case, the birth and development of IPUs have become logical.

Elf dance: IPU presents the change of the times for the data center

The data center must complete the new upgrade and evolution at this stage, especially for new computing institutions. In the future, the data center needs to achieve a large -scale distributed heterogeneous computing collaborative work and seamless connection. In particular, the calculation unit for the acceleration of infrastructure can effectively load the work of the CPU, which is essential for the improvement of data centers.

In 2021, Intel launched an infrastructure processor (IPU). As a programmable network device, Intel IPU can help cloud and communication service providers, as well as enterprises to improve security, save expenses, and fully release the performance of the central processor (CPU). With the Intel IPU platform, the data center can better release computing resources through security, stable, and programmable solutions, and achieve higher security and isolation. IPU, a small elves that help data centers to release computing power, can bring a number of advantages to microservices innovation of cloud data centers. As a new product form. Intel IPU can also bring higher levels of security and control to future cloud data centers while covering smart network card functions. Its advantages include:

1. Reduce the overhead of the virtual machine management program and infrastructure stack in the host -side processor, so that more processor resources can be used for application and tenant workloads.

2. Uninstall the storage stack from the host side processor, so as to release more processor resources for applications and tenants' workloads.

3. Uninstall high -dense infrastructure tasks, such as encryption and decryption and packet processing.

4. Under extreme circumstances, Intel IPU can uninstall the entire virtual machine management program and release all core kernels to support the application and microservices of the processor, which is essential for naked metal service products.

5. The possibility of uniform cloud management and control for nude metals and cloud hosting.

From the perspective of technical architecture, Intel IPU can bring a series of development advantages to cloud data centers, because data centers can effectively uninstall the network, storage, security, and infrastructure management from processors to Intel IPU, so as Realize computing power release and accelerate multiple infrastructure capabilities. These capabilities include--

1. Network acceleration: Uninterled virtual switch software that carries network I/O, data forwarding and other functions, such as OVS, etc., uninstall from the host side processor to the IPU to enhance the network throughput and reduce the delay of network processing.

2. Storage acceleration: Transfer the storage interface and protocol stack of Virtio-BLK, NVMe-OF and other storage interfaces from the host side processor to the IPU, increase storage elasticity and flexibility, and reduce system complexity and expenses.

3. Safety acceleration: Uninstall a large amount of encryption/decryption, compression and other security functions from the host side processor.

4. Infrastructure processing: Uninstall the cloud service management function from the host side processor to the Intel IPU, making the allocation and management of virtual machines, containers or nude metal services more efficient.

In addition to the introduction of stronger and flexible and programmable hardware to optimize acceleration for specific functions, release more

In addition to valuable computing power, Intel also gave IPU products a rich software ecosystem. At present, Intel is accelerating the rapid development and deployment of cloud data center performance acceleration solutions through the construction of a better ecosystem. In the planning, the ADP platform will provide users with a board card hardware design, software, driver, and technical design support implemented through the Intel Open FPGA stack (Intel OFS). Users can use these software tools and technical support to accelerate the development of IPU -related products, and quickly push the products to the market.

In addition, it should be noted that the processor is the heart of the data center. Intel IPU into the cloud data center architecture can effectively use the system capabilities and platformization value between it and the processor. By connecting the Intel Solidarity processor, Intel IPU can effectively enhance the computing power of cloud data center and fully release heterogeneous ingenuity.

In the efficient energy cloud data center ecological solution provided by Intel for users, not only can it be uninstalled through the IPU and accelerate the performance of infrastructure, but also introduces the Intel Salvable Expansion processor to further strengthen the computing power. Through the "open source throttling" of computing power, faster data processing speed, larger bandwidth access ability, and lower network delay.

The third -generation Intel Solidarity processor can bring the performance of the server to enhance the performance::

1. More kernels and better architectures bring greatly improved computing power performance, which can effectively respond to high -density calculations;

2. Support more memory, support PCIE-GEN4, can achieve higher I/O bandwidth per nuclear I/O;

3. Multiple built -in acceleration technology, such as Intel deep learning acceleration technology (Intel Boost), can provide strong acceleration capabilities in artificial intelligence and other scenarios.

Another advantage of introducing Intel Xeon Expansion processors in the computing power architecture is that it can effectively improve the efficiency and smoothness of uninstallation capacity. Because the host side and IPU use the Intel -based processor, it can easily migrate the host's side operation to the IPU, or even the need to migrate without compiling, greatly improving the efficiency of development, deployment and testing.

Under a series of values ​​brought by IPU, cloud data centers can continue to move towards the era of microservices innovation and open the road of future data centers.

Help Qiaoqi: Baidu Smart Cloud's Pioneer Explorer

In the white paper, we can see the cooperation cases and industry demonstration value of Intel IPU and Baidu Smart Cloud. It can be seen from the cooperation between the two parties that Baidu Smart Cloud has successfully achieved the value of improvement of server efficiency, improving the level of security, and reducing the value of comprehensive costs. In the tide of digital intelligence brought by cloud and AI, cloud services continue to extend to various fields and carry more and more core business capabilities. This also leads to the challenges of cloud computing service providers that need stronger computing power resources, larger bandwidth access and lower network delay. At the same time, cloud service brands also need to gain value from the innovation of computing structure to ensure their own competitive advantages. To this end, Baidu Intelligent Cloud proposed a cloud and wisdom strategy to vigorously develop the AI ​​cloud business. The computing power requirements index of AI cloud services increase. Facing the needs of the future of the Yuan universe, the release of computing power will also face more challenges.

In this process, Baidu Smart Cloud took the first time that the efficient software and hardware programmed by Intel FPGA and Intel to the Geority D processor launched the self -developed DPU device -Baidu Taihang DPU 1.0 for the first time. It supports Taihang elastic nude metal services, including network uninstallation and PCIE device hot insertion and other functions, so that the virtual overhead of cloud computing is largely precipitated on the network card, which greatly improves the performance of cloud products. At present, Taihang DPU 1.0 products have been applied in a variety of cloud service scenarios such as Taihang elastic nude metals, and have received good evaluations of users. Practice has proved that Baidu's independently developed Baidu Taihang DPU1.0 can effectively solve the limitations of cloud data center management and virtualization.

Baidu Taihang DPU1.0 products are equipped with 2 25G bandwidth light mouths, and introduced Intel's FPGA and Intel to strong D processors as the core of computing power to realize the unloading of various virtualization functions. It provides users with elastic network card In addition to elastic storage capabilities, it also supports hot insertion of 1024 equipment. At the same time, by uninstalling the plane of the cloud tube control to Baidu Taihang DPU 1.0, the virtual cloud host and naked metal co -pool can be achieved, that is, nude metals can be uniformly scheduled with virtual cloud hosts. Test verification from Baidu Smart Cloud shows that Baidu Taihang DPU1.0 products can provide users with a forwarding rate of 10 million PPS and the storage performance of 200,000 IOPS.

Facing the future, DPU products will also become the core components of smart cloud services, helping the uniform flexible base of IaaS resources, and then provide reliable guarantees for users to carry out various business innovation and implement industrial intelligence transformation relying on Baidu Smart Cloud.

The architecture innovation brought by the IPU can ensure the maximum computing power release from the base layer and release the core competitive advantage of the data center. Since 2016, "the second half of the Internet" has become a proposition for extensive discussion. Nowadays, at the end of large -scale competition, technological innovation and refined services have become the stage of the main competition point, and "cloud computing the second half" seems to have arrived.

In this industrial cycle, the competitiveness of cloud computing service providers is not only in the cost -effectiveness of infrastructure, but also to provide customers with service innovation, technological innovation, and comprehensive value. In this case, the infrastructure innovation of the cloud computing data center has become crucial. It is not only a guarantee of intensive computing power, but also the cornerstone of business innovation.

With the Intel IPU and related products based on its reference design and development, in the innovation innovation of the cloud data center's infrastructure and market value, and gradually forming a sustainable ecological chain, Intel and Baidu Smart Cloud and other in -depth cooperation The partners are also focusing on the future development direction of cloud services to open up more optimization solutions from the software and hardware level. For example, a larger network throughput is provided through the new Intel IPU platform to achieve a comprehensive decoupling of hardware resources, and it is extremely low delayed.

The current to the end to accelerate.

Facing the future, Intel plans to continue to make efforts in the IPU product system. Intel launched two second -generation 200G programmable IPUs in 2022, codenamed "Oak Springs Canyon" and "Mount Evans". Among them, Oak SpringsCanyon is equipped with Intel -Qiangqi D processor and Intel Agilex ™ FPGA chip. Mount Evans is Intel's first IPU product based on ASIC chip. Both products will have the infrastructure acceleration capabilities required to uninstall the 200 G workload, and both can work with the Intel Xeon Expansion processor. In the next two to three years, Intel will launch an IPU product that supports 400 G, and in 2025 or later, a new generation of 800G IPU products based on FPGA chips and ASIC chips.

At the same time, Intel is also further relying on rich open source software (such as Internal Package and Development Kit, IPDK, SPDK, DPDK, etc.), dedicatedCollaborative and positive community interactions allow Baidu and other cloud service providers, cloud service end users, and other partners to accelerate the construction of IPU -related solutions, and capture greater value from the continuous innovation of cloud data center infrastructure.Future data centers are accelerating, and IPUs are becoming a dancing elf.

- END -