Keysight Presents at AI Infrastructure Field Day 6

The Design to Deployment Continuum Validating Al Infrastructure at Scale with Keysight

Watch on YouTube
Watch on Vimeo

Venkat Pullela, CTO for AI infrastructure at Keysight, addresses the critical challenge of validating large AI clusters for capacity, capability, and deployment readiness. Pullela notes that the AI era introduces a unique problem where distributed, loosely coupled systems must operate synchronously, a departure from traditional multi-processor or hyperscale architectures. This demands a fundamentally new approach to ensuring the reliability and performance of AI infrastructure.

A significant challenge in AI infrastructure is the “weakest link” failure, where the performance or failure of a single component can impact the entire synchronous system, making quality a top priority. Unlike previous eras, AI workloads run “full throttle” constantly, meaning the slowest packet or link (the “long tail”) determines overall system performance, rather than an average. This requires AI systems to be designed and mass-produced with the precision and systemic integration of a Formula One car, prioritizing performance as the primary criterion. Keysight’s solution, the “Design to Deployment Continuum,” aims to meet these demands by accelerating time-to-market, enhancing quality through systems thinking, and optimizing efficiency via workload-centric designs.

Keysight’s continuum approach enables system testing from “day zero,” allowing validation to commence before physical hardware or even all code exists. This is accomplished by creating a unified system testbed that seamlessly integrates simulated, emulated, and real components. This allows for early identification and mitigation of high-risk design decisions. The company provides tools and frameworks that maintain a uniform workflow, common models (for infrastructure, applications, and user behavior), and standardized benchmarks across the entire lifecycle: from initial design and chip development to rack integration and full data center deployment. Keysight positions itself to “be the rest of the world” for any component under test, facilitating continuous testing, co-design, and fine-tuning of AI infrastructure for diverse and specific application requirements, including end-to-end network performance.

Personnel: Venkat Pullela

Building Sovereign Al and Neoclouds From Components to Tokens with Keysight

Watch on YouTube
Watch on Vimeo

Venkat Pullela, CTO of AI Infrastructure at Keysight, addresses the interesting topic of building AI infrastructure amidst geopolitical changes, focusing on the concept of sovereignty, where entities aim to build their own systems. The presentation covers what this new definition of sovereignty means and how hyperscalers, sovereign nations, and neo-clouds operate in this space, developing solutions from low-level components to higher-level abstractions like tokens. Pullela’s role involves maintaining line of sight to rapidly evolving technology and driving proofs of concept into products.

The AI infrastructure is a vast and complex ecosystem, comprising three sub-ecosystems: infrastructure builders, data center/telecom operators, and edge infrastructure/endpoint users. Building this entire system independently is challenging due to the sheer number of components and rapid technological changes. Keysight proposes solutions centered on “chiplets” and “global talent.” Chiplets offer modularity and freedom in design, allowing entities to specialize while relying on partners for other parts. Global talent acknowledges the necessity of international collaboration for expertise. This approach elevates the importance of rigorous conformance testing, open standards, and industry-wide co-design to manage the integration complexity and ensure interoperability across diverse components and vendors.

Keysight’s core contribution is providing a “design-to-deployment continuum” to ensure a smooth time to market. This involves mastering numerous technologies, from pre-silicon validation, photonic design automation, and wafer testing, to protocol standards and conformance. Keysight offers a range of tools and solutions that span this entire lifecycle, enabling systematic testing, vendor benchmarking, fabric benchmarking, and the emulation of complex environments, such as thousands of cell phones or RF environmental simulations, to evaluate AI token rates and other performance metrics. The goal is to provide common APIs, models, and conformance testing criteria that apply from simulation to real-world deployment, allowing hyperscalers, sovereign countries, and neo-clouds to efficiently build and control their AI infrastructure despite the inherent complexity and rapid evolution.

Personnel: Venkat Pullela

Validating Large Al Cluster Fabrics Using Keysight Al Data Center Builder

Watch on YouTube
Watch on Vimeo

Keysight aims to help organizations build sovereign AI and neoclouds by addressing critical challenges in validating large AI clusters and their underlying fabrics. The presentation highlights significant inefficiencies observed in AI training tasks, where over 43% fail, with half of these failures (21%) directly attributable to network issues. Such failures lead to expensive checkpoint reloads and idle GPU time, impacting return on investment. Furthermore, even in successful tasks, GPUs spend more than half their time waiting on communication, indicating low utilization and prolonged training times, with only 20% of the time actually dedicated to computation.

To tackle these problems, Keysight introduces its flagship product, Keysight AI Data Center Builder, designed to benchmark, optimize, and co-tune AI fabrics before they are deployed in production. This tool offers three primary applications: collective benchmarks, which emulate single AI collective operations (data movements between GPUs) to provide metrics like network utilization, job completion time, and congestion; a mix of collectives, which takes it further by simultaneously emulating several collective operations across different nodes to assess fabric behavior under complex loads; and workload emulation, utilizing “chakra traces” for sequential model training steps, facilitating regression testing and training time optimization.

The Keysight AI Data Center Builder is available in both hardware (e.g., Aries One chassis) and software versions, and it provides comprehensive insights into fabric performance. It offers key metrics such as collective completion time, algorithm bandwidth, bus bandwidth, and ideal performance percentages, along with duty cycle to indicate GPU utilization. Visualizations like cumulative distribution function (CDF) graphs help identify tail latency, while monitoring Priority-Based Flow Control (PFC) frames indicates network congestion. The tool also provides timelines and heat maps to pinpoint performance outliers and issues like reordered packets. By enabling users to test and optimize their fabric without engaging expensive GPUs, Keysight ensures that the underlying infrastructure is robust and highly efficient, thereby increasing GPU utilization, reducing idle time, and ultimately making AI training jobs complete faster and more reliably.

Personnel: Jess Tarkar