Qualification Test Report for AI Data Center Network

This technical report presents the qualification test results for the AI Data Center Network utilizing Juniper Apstra, NVIDIA GPUs, and WEKA Storage within the Juniper Validated Design (JVD) framework. The evaluation focuses on the AI ML Cluster solution, specifically examining the IP Clos deployment aimed at enhancing congestion control. The report outlines the test topology, which includes various components such as QFX5220 and QFX5240 switches, NVIDIA GPU servers, and the orchestration by Juniper Apstra. It details the testing approach that emphasizes validating congestion control scenarios using Junos 23.4X100-D20, with specific goals to optimize performance metrics like Job Completion Time (JCT) and throughput. The document further discusses the parameters fine-tuned during testing, including shared buffer allocation and PFC watchdog settings. Additionally, it includes performance data from multiple test scenarios, comparing JCT values against MLCommons benchmarks, and highlights features and events tested within the JVD framework.