How to Test AI Data Center Networks

+ 800GE Traffic Emulator

Reproducing Realistic Network Behavior of AI Workloads

Benchmarking artificial intelligence / machine learning (AI / ML) networks with realistic workloads typically requires investments into computing systems with GPUs and remote direct memory access (RDMA) network interface controllers (NICs), which are costly to build and time-consuming to operate. Deploying and operating these systems for terabit-scale validation and experimentation in the lab is necessary to fully optimize the AI networks. Proper benchmarking and testing of AI networks requires analyzing various parameters such as fabric topology, congestion control algorithms, load balancing mechanisms, or retransmission capabilities.

Generating realistic, high-scale AI workload traffic for network benchmarking requires RDMA over Converged Ethernet (RoCEv2) endpoint emulators and software with prepackaged methodologies that support collective communications patterns (including all-to-all, all-reduce, all-gather, and more). The software provides workloads specific to AI networks which help measure key metrics such as job completion time and bus bandwidth. More advanced validation methodologies include emulation of multiple tenants with mixed parallel collectives, as well as replay of predefined AI workload traces.

AI Data Center Network Test Solution

Testing an AI data center network requires network traffic emulators and software with prepackaged methodologies that support AI workloads. The AI data center network test solution includes Keysight AresONE 1600GE RoCEv2 endpoint emulation paired with Keysight AI Data Center Builder software. This solution can repeatably create scenarios with different data sizes resulting from collective communications in an AI cluster. Each port on the AresONE emulates GPUs and RDMA NICs. The traffic includes emulating queue pair (QP) connections and flows, generating congestion notifications, performing Data Center Quantized Congestion Notification-based (DCQCN) dynamic rate control, and providing flexibility to test throughput, buffer management, and equal cost multi-path (ECMP) hashing. With this solution, engineers can perform multiple design experiments, measure the performance improvements in the lab, and then apply the results to a production environment without the need for dedicated AI compute nodes and NICs in the lab.

Get Quote

See Demo of Our AI Data Center Test Solution

Explore Products in Our Network Testing Solutions

941-0134 AresONE 1600GE-C, OSFP1600 IHS

Learn More View Data Sheet
950-2001 KAI Data Center Builder Start Bundle with Collective Benchmarks Application

Learn More View Data Sheet
930-2208 IxNetwork RoCEv2 Lossless Ethernet Test Package

Learn More View Data Sheet