Two data center engineers review information on a laptop while standing between rows of illuminated server racks.

Advancing Lab and Test Automation for AI Workloads

AI is having an outsized impact on networks, with workloads driving more uplink traffic and volumes increasing dramatically.

Data-intensive AI workloads must be processed in near real-time and provide high-bandwidth, low-latency, and lossless connectivity. This high-volume, high-speed AI traffic is pushing network infrastructures to support continuous and dense utilization at scale while mitigating the harsh thermal conditions generated.

At the same time, quality of service (QoS) expectations and service level agreements (SLAs) demand these complex infrastructures handle workloads without failure.

The bottom line: AI data centers must satisfy strict performance requirements at massive scale, while reducing power and heat. Optical switching at the physical, fiber layer and optical circuit switches (OCS) are helping by delivering significant cost savings, reduced power consumption, and improved latency.

The value of optical circuit switches

An optical circuit switch enables uninterrupted, high-bandwidth data transmission by creating a dedicated optical path between nodes in a fiber infrastructure. A fully optical path eliminates the need for conversions between electronic and optical signals. These conversions are power-hungry and add latency to the end-to-end path. As a result, OCSs efficiently handle high-bandwidth AI data streams by maximizing speed and minimizing signal loss, latency, and power consumption.

Highly scalable, all-optical networks are essential for meeting the growing demands of AI, including graphics processing unit (GPU) scale-up/scale-out and data center interconnect. Now, OCSs are emerging to connect directly to GPUs via optical interfaces on the system-on-a-chip (SoC), optimizing GPU-to-GPU communications.

AI labs ensure performance

AI networks are extremely complex, with stringent performance requirements and innovative new technologies. Testing them in the lab is equally complex.

Manual processes are not feasible for AI labs, where speed, scale, and dynamic workloads prevail. The large number of AI connections and test configurations, as well as the volume of diverse AI traffic, make test and lab automation and resource reuse essential.

AI test labs must scale up, scale out, and dynamically reconfigure GPUs based on anticipated AI workloads. Hyperscalers and telcos are validating AI models and infrastructure in staged, controlled labs before entering the production environment, subjecting them to realistic environments across the lab to live deployment lifecycle. Now these test labs must scale up and out while dynamically reconfiguring GPUs based on anticipated workloads, with ongoing monitoring and optimization of utilization.

Robotic optical cross-connects

To address the scalability and latency needs of AI networks, Telescent has introduced an automated, configurable optical cross-connect, the G5 Network Topology Manager (G5 NTM). The G5 NTM achieves physical layer automation with software-driven physical network topology management that enables efficient and reliable network operations for networks with high complexity and scale.

One G5 NTM rack unit can connect and configure 1,000 single fibers or jackets of 16 fibers, so a total of 16,000 fibers is possible (80 times more than prior solutions). The all-optical patch-panel is controlled by software to provide dynamic reconfiguration of optical circuits. Robotic arms enable connections that are ultra-fast with low insertion loss, eliminating human error. Algorithms enable fibers within an enclosure to be routed and provisioned without becoming tangled — a key benefit over manual configurations.

G5 NTMs are already deployed in production networks. Their massive level of scale and density is a preview of the radical impact AI/machine learning (ML) workloads have on data centers and today’s networks.

Lab and test automation for AI scale

AI test labs need optical connectivity, automation, and software-controlled test scripts to exercise the G5 NTM as well as other optical cross-connects and switches. And to test at a volume and speed consistent with the G5 NTM. This reality makes lab and test automation unavoidable:

Lab automation enables teams to consolidate and manage multiple labs as a single, web-accessible resource. All lab resources (servicers, GPUs, test equipment, analyzers, signal generators, etc.) are attached to the Telescent G5 and then cross-connected on-demand with each test, staging, and production deployment. Lab automation accelerates testing, eliminates manual effort, maximizes resource utilization, and reduces test equipment capital expenditures.
Test automation infrastructure creates test beds programmatically, instantiates the test bed, and automates the connections, including interfacing with Telescent’s software to manage thousands of connections. Leading vendors and operators are incorporating continuous, automated test (CT) into their continuous integration and continuous delivery (CI/CD) processes, so tests can be rapidly set up and torn down.

Keysight partners with Telescent to accelerate lab and test automation for AI

Keysight’s strategic partnership with Telescent integrates Telescent's optical circuit switch and high-density smart patch panel technologies and software with Keysight’s Velocity test lab automation portfolio.

The integration creates a powerful force multiplier for companies seeking to optimize AI/ML resource connectivity, automate lab and test infrastructures, and maximize return on investment.

With its partnership, Keysight’s Velocity test automation portfolio integrates Telescent's advanced optical switching technology, further advancing test automation and AI/ML resource optimization to enable scaled optical connectivity in AI labs and data centers.

The solution drives lab automation to improve operational efficiency and resource utilization through resource sharing, provisioning, and orchestration via a single cloud-based user interface. For example, Velocity automatically controls and inventories any-to-any optical switch connections. Test environments can also be saved and shared across teams for test reproducibility.

The combined solution delivers significant value to companies automating their test infrastructure, such as enterprise network test labs, AI/ML labs, data centers, and equipment manufacturer validation facilities seeking to improve operational efficiency.

Keysight’s innovative, automated lab and test solutions help companies ensure reliability, security, and performance in their operational networks. The solutions include software and services, implementation expertise, testing methodologies, and a robust switching infrastructure that continues to grow as new technologies arise.