Spirent Communications, a provider of test and assurance solutions for next-gen devices and networks, has introduced a high-density test solution capable of emulating realistic AI workloads over Ethernet.
Spirent’s AI test solution, operating on the A1 400G platform, can emulate high-density 400G xPU workloads for AI environments, enabling customers to assess their Ethernet fabrics without the need for a new xPU server-equipped labs.
Ethernet serves as the primary technology for networking. Testing realistic AI traffic workloads and their impact on AI data center networks is important but evaluating the performance of Ethernet fabric in new AI environments has proved to be challenging, due to unique AI training requirements such as increased workload volumes and sensitivity to latency and congestion.
Leveraging the RoCEv2 protocol, the Spirent platform has been designed for ease of use, to be simple to configure, and to deliver consistent results, simplifying the testing of AI use cases. It can also function as a platform capable of concurrently testing AI and routing/switching use cases.
“AI is an important focus for Spirent as applications such as ChatGPT, Lensa, Copilot and others are changing the communications landscape, and hyperscale cloud providers shift investment from their traditional data center front-end focus to new back-end infrastructures needed to manage the explosion in AI applications and workloads,” said Eric Updyke, CEO of Spirent Communications.
“These new environments are increasingly being built and operated separately from traditional data centers and are physically very different in order to cope with the specific needs of AI.”
“A top priority for the Spirent team has been to develop a dedicated test solution capable of emulating realistic xPU workloads and AI traffic patterns with ease. Our new solution will enable engineers to test their Ethernet fabric without having to go to the expense of building a whole new lab of costly xPU servers and configure test cases to generate AI workloads using these real servers.”
