VIAVI has launched an Ultra Ethernet Transport (UET) validation solution for AI fabrics, built as a GPU-free offering for the VIAVI TestCenter platform. The goal is to validate next-generation high-speed AI networks, including both scale-out and scale-up AI back-end networks, using the Ultra Ethernet Consortium’s Ultra Ethernet Stack.
The solution emulates the transport layer of UET and is aimed at reproducing realistic, stateful AI traffic patterns at scale without deploying dedicated GPU infrastructure. VIAVI lists UET behaviors covered by the emulation as reliable ordered and unordered delivery (ROD/RUD), packet trimming, congestion control, and dynamic multipathing.
VIAVI also says the platform supports full-fidelity emulation of AI workloads, including collective communications (CCL) and large language model (LLM) flows. For fabric-level behavior, it includes validation of load-balancing mechanisms such as ECMP, packet spraying, and flowlet switching.
For data center engineers building AI fabrics, the practical value is straightforward: validating transport and congestion behavior is hard to do with “real” workloads at meaningful scale, and burning expensive accelerator capacity just to generate test traffic doesn’t pencil out for many teams. A test rig that can reproduce UET transport behaviors deterministically can shorten bring-up cycles, especially when you’re trying to baseline multi-vendor behavior and tune congestion control before large deployments.
“AI clusters will soon scale to millions of endpoints, which means relying on physical GPUs alone to validate network behavior is no longer practical,” said Aniket Khosla, vice president of product management, Optical Transport and High-Speed Ethernet, VIAVI. He added that the “GPU-free, full-fidelity UET validation solution gives customers the confidence to deploy scalable, high-performance AI fabrics faster and more cost-effectively.”
HPE said it has been collaborating with VIAVI on UET transport validation, including work using the Juniper QFX5240 platform and Junos Evolved, and noted advanced software features such as packet trimming. “The VIAVI TestCenter platform provides realistic traffic emulation and the granular visibility needed to validate congestion control and transport performance at scale,” said Mahesh Subramaniam, senior director of product management, AI Data Centers, HPE.
Source: VIAVI
