In scenarios such as distributed storage, high-performance computing (HPC), and AI, RDMA over Converged Ethernet version 2 (RoCEv2) is used to reduce CPU processing workload and delay and improve application performance. Distributed high-performance applications use the N:1 incast traffic model. For Ethernet switches, incast traffic may cause instantaneous burst congestion or even packet loss in the internal queue buffer of a switch. As a result, the application delay increases and the throughput decreases, which lowers the performance of distributed applications.
An intelligent lossless network uses the AI-ready hardware architecture and iLossless algorithm — an AI-powered intelligent lossless algorithm — to achieve the maximum throughput and minimum latency without packet loss in AI, distributed storage, and HPC scenarios. This accelerates computing and storage efficiency and builds a converged network for future DCs.
What Are the Advantages of an Intelligent Lossless Network?
AlphaGo's victory in March 2016 was a major milestone in AI research, declaring that the Fourth Industrial Revolution characterized by AI is coming. More and more enterprises incorporate AI into their subsequent strategies for digital transformation. Against this backdrop, the focus of enterprise DCs in the AI era is shifting from fast service provisioning to efficient data processing. Computing, storage, and network are the three key elements of a DC, which promote each other. With the explosive growth of AI application data, heterogeneous computing based on general process units (GPUs) and AI chips is booming, and the computing performance has been improved by 600 times in the past five years. In the data storage field, the access performance of SSDs is 100 times higher than that of HDDs, and the performance of Non-Volatile Memory Express (NVMe) is 100 times higher than that of SSDs. The rapid development of computing and storage poses new requirements for data center networks: zero packet loss, high throughput, and low latency. In response to such requirements, intelligent lossless networks emerge.
The intelligent lossless network solution has the following visions:
Fully converged Ethernet network
Leveraging RoCEv2, the intelligent lossless network aims to implement Ethernet-based unified management of the service network, computing network, and storage network. This solves the problem of multiple networks using multiple technologies in the traditional data center network in the IP and FC eras, and meets the low latency requirements of the data center storage network.
Fully converged Ethernet networkApplication acceleration
The intelligent lossless network provides intelligent lossless algorithms such as Convolutional Neural Network (CNN) and Deep Q-Network (DQN) to resolve network congestion and offload applications to the network, accelerating computing and storage applications.
Autonomous driving network
The intelligent lossless network uses intelligent lossless algorithms such as CNN and DQN to automatically learn network parameters, and adjusts the network accordingly to achieve zero packet loss, maximum throughput, and minimum latency.
