As the digital base of the information society, data centers have become a strategic infrastructure to promote social development. Today, in an environment based entirely on Zero Trust, data centers require unlimited computing power and high security performance. "To deal with an environment like this, the best technology right now is cloud-native." Song Qingchun of NVIDIA's network business unit said.
Quantum-2 platform that will be calculated
Earlier last month, NVIDIA released its next-generation InfiniBand web platform, the NVIDIA Quantum-2 platform. According to reports, NVIDIA Quantum-2 will provide cloud computing providers and supercomputing centers with extreme performance, extensive access capabilities and strong security.
"The launch of the Quantum-2 platform is based on the current state of explosive growth in data and computing volumes. Song Qingchun said, "The traditional demand for cloud services is mainly flexible and convenient, but when large computing resources are needed, it is often necessary to use the resources of the supercomputing center." How to integrate the performance of supercomputing with the flexibility and security of the cloud to provide an updated architecture to support the existing growing business demand for computing power? Song Qingchun stressed: "It is necessary to bring supercomputing technology into the data center through cloud-native supercomputing technology, so that the data center has supercomputing technology at the same time, but also has flexibility and security." Cloud native will be the future trend of providing computing power platforms. ”
According to reports, the NVIDIA Quantum-2 platform is the 400Gbps InfiniBand network platform, including the NVIDIA Quantum-2 switch, ConnectX-7 network card, BlueField-3 InfiniBand DPU (data processor) and all the software that supports this new architecture. With its cloud-native technology, NVIDIA Quantum-2 delivers high throughput of 400Gb/s per second and advanced multi-tenant support to meet the needs of many users.
It is reported that when supercomputers and cloud-native supercomputing systems want to achieve high performance, all resources need to be involved in the computing. The goal of Quantum-2 is to realize that where the data is, the computation is there. Based on the above advantages, Quantum-2 can help enterprises realize a technology that is critical to AI business - In-Network Computing.
Gilad Shainer, senior vice president of NVIDIA's Networking Division, said: "Today, the demands of supercomputing centers and public clouds are converging – they must provide the highest possible performance for the next generation of high-energy computing (HPC), AI and data analytics applications, while also securely isolating applications and responding to different user demands for traffic. With NVIDIA's Quantum-2 InfiniBand platform, modern data centers can already make this vision a reality. ”
3U in one empowers unlimited computing power
With the improvement of computing power, especially the development of heterogeneous computing, in recent years, GPU-accelerated computing has become the key to artificial intelligence, deep learning and big data analysis applications. In the era of big data, information and data show an explosive growth trend, and DPU has become the third computing unit in addition to the CPU and GPU in the accelerated computing model of data center.
At the NVIDIA GTC 2021 Summit, NVIDIA officially released the NVIDIA BlueField-3 DPU, a next-generation data processor that provides powerful software-defined networking, storage, and network security acceleration capabilities for data centers. NVIDIA's BlueField DPU lays the foundation for a 3U all-in-one architecture.
According to reports, after using blueField DPU, all CPU resources can be used to run business applications, and part of the CPU resources are no longer needed to support SDN. In addition to this, it can also increase the ability to scale, cloud-native applications are highly distributed applications that bring dense "east-west" traffic within the data center, and BlueField DPUs provide a high-throughput, low-latency network environment for scaling applications.
In addition, multi-tenancy and infrastructure resiliency in cloud data centers pose privacy and confidentiality risks that BlueField DPUs deal with well. BlueField DPUs also provide a robust and robust network to address the needs of a wide range of applications where GPU-accelerated computing is growing in the cloud, enterprise, and edge.
Song Qingchun said: "The emergence of DPU has made up for the lack of infrastructure acceleration capabilities in data centers, and realized a new data center architecture with DPU, GPU and CPU 3U, making data centers a new computing unit. When giving us a way to optimize computing power resources, the 3U integration has become an inevitable architecture of the data center from the higher level, higher granularity, and larger space of the data center. ”
Zero Trust Challenge
At present, enterprises are facing an increasing amount of data and faster and faster data transmission. All kinds of heterogeneous data appear frequently, the amount of unstructured data is much larger than that of structured data, and the complexity of processing unstructured data is much greater than that of processing structured data. Coupled with the need to do cross-platform integration, various model iterations, computing iterations, platform iterations are getting faster and faster, and network security has become the biggest threat to data security.
How do I resolve this issue? To help businesses address increasingly sophisticated threats, NVIDIA has released a Zero Trust cybersecurity platform. Leveraging the platform's suite of capabilities, the cybersecurity industry can build solutions that protect enterprise data centers in real time. The Zero Trust Security Platform combines three technologies – NVIDIA BlueField DPU, NVIDIA DOCA, and NVIDIA Morpheus Cybersecurity AI Framework.
According to reports, NVIDIA DOCA 1.2 is a special SDK for the zero-trust security framework, in DOCA1.2 provides APIs for Load Balancers, DPI, IPS, IDS, next-generation firewalls and other security services, users of security software design or security software vendors, can directly call the hardware acceleration engine in the DPU through the DOCA API to make the data center more secure.
In addition, NVIDIA Morpheus is a deep learning cybersecurity framework for NVIDIA in the DPU+DOCA Secure Data Center. The latest release includes a new workflow that uses unsupervised learning to create digital fingerprints to detect when a cyber intruder takes over a user account or machine. NVIDIA Morpheus accelerates data science processing, generates pre-trained or inference models from NVIDIA AI training, and distributes new Policies to NVIDIA DPUs, enabling enterprises to simultaneously inspect all telemetry data in their data center networks.
Song Qingchun pointed out: "The demand for computing power in data centers is becoming larger and larger, large-scale models are emerging in an endless stream, and it takes a lot of GPUs to run large models to process in parallel, and the simultaneous use of data parallelism and model parallelism in future work training will become a trend." Under such large-scale use scenarios, how to maintain a high computing power, but also ensure training performance and ensure service security under multi-tenant conditions? Cloud-native technologies are essential for such a trend. ”