Calnex Solutions says it has demonstrated 400G Ethernet network emulation at full line rate with 100 percent throughput while allowing Priority Flow Control (PFC) frames to pass through transparently. The company is targeting validation testing for lossless Ethernet fabrics used in AI clusters, where engineers need to reproduce congestion and latency effects in a controlled lab environment.
The capability is aimed at testing and optimizing lossless Ethernet fabrics that use PFC, including long-haul data center interconnect (DCI) links. Calnex’s stated approach is to introduce impairments such as congestion and delay without intercepting control traffic, which it says preserves the behavior and semantics of lossless Ethernet during validation.
For data center engineers building large-scale AI clusters, the practical issue is that network pathologies are often intermittent and workload-dependent. Microbursts, congestion, and PFC interactions can create bottlenecks and drive up tail latency, which in turn can reduce GPU utilization and stretch job completion times. If a lab setup can’t recreate those conditions consistently, it’s hard to qualify changes in switches, optics, cabling paths, or fabric tuning with much confidence.
“In large-scale distributed compute environments such as training clusters, network inefficiencies can leave a significant share of GPU capacity underutilized,” said Tommy Cook, CEO and Founder at Calnex Solutions. “Our breakthrough is the ability to recreate these conditions at full throughput while preserving the native behavior of lossless Ethernet, helping engineering teams test performance and reduce deployment risk.”
Calnex said the milestone supports its AI validation work across inference, AI-RAN, and data center networking.
Source: Calnex Solutions
