Huawei CloudEngine Data Center Switches

CloudEngine Builds Intelligent and Lossless Data Center Networks for the AI Era

• 
• Embedded AI Chip, Ensuring Zero Packet Loss

  The CloudEngine 16800 provides high-density 400GE line cards with a switching capability of up to five times the industry average. This meets traffic multiplication’s requirements in the AI era. With the embedded AI chip, CloudEngine switches can intelligently analyze network traffic models and implement zero packet loss, low latency, and high throughput. The AI Fabric intelligent and lossless data center network solution composed of CloudEngine switches has passed the EANTC’s high-performance data center test. Latency can be reduced by 44.3 percent.

  Before

  After

  Latency can be reduced by 44.3 %

  Read EANTC Test Report Watch EANTC Test Video

• 
• Unique Intelligent Algorithm Set, 30% Higher Efficiency than Industry Average

  Huawei CloudEngine data center switches use the iLossless intelligent algorithm. The switches implement global optimization based on dynamic Explicit Congestion Notification (ECN), Virtual Input Queue (VIQ), and fast Congestion Notification Packet (CNP). They achieve the highest network performance and optimal application experience. In a competition test conducted by the Tolly Group, the AI Fabric intelligent and lossless data center network solution constructed by CloudEngine switches leads in High-Performance Computing (HPC), distributed AI training, and distributed storage scenarios.

  Before

  After

  Efficiency compared to industry 30 %

  Read Tolly Test Report Watch Tolly Test Video

• 

  
• Convergence of Computing, Storage, and Data Networks Reduces TCO by 53 Percent

  Three networks need to be deployed in a traditional data center: HPC network based on InfiniBand (IB), storage network (SAN) based on Fibre Channel (FC), and LAN (switching network) based on Ethernet architecture. The first two dedicated networks need professional O&M staff, which is costly. Huawei CloudEngine switches use the embedded AI chip and algorithm to build a converged Ethernet network with zero packet loss that reaches 100 percent computing power. They implement convergence of three networks and reduce the TCO by 53 percent.

  Before

  After

  TCO can be reduced by 53 %