What's Happening?
Credo Technology Group Holding Ltd (Credo) and Oracle have announced that the Open Compute Project (OCP) Community has approved two new specifications aimed at enhancing telemetry in optical transceivers.
These specifications were developed through the OCP Networking Project's Optics Reliability Workstream, co-led by Chris Collins of Credo and Steve Manley of Oracle. The primary goal is to establish a common, standardized framework for optical telemetry, which will facilitate faster deployment and more reliable operation of AI fabrics by minimizing network disruptions. The approved specifications define an open, interoperable standard for optical transceivers and hosts to coordinate out-of-band communication, enabling them to share Layer 1 (L1) telemetry, compute a common link-health metric, log significant link events, and support automated responses to prevent and remediate link issues.
Why It's Important?
The standardization of optical transceiver telemetry is crucial for the rapidly expanding AI infrastructure, which relies on hundreds of thousands of optical links. Even brief link disruptions in such large-scale systems can significantly impede AI training runs and degrade overall cluster performance and returns. By establishing a common software interface for enhanced telemetry, Credo and Oracle, through the OCP Community, are addressing a critical need for consistent operational visibility and reliability across different vendors and form factors. This initiative will empower operators with the necessary diagnostic information to proactively manage and resolve optical link issues, thereby improving the efficiency and uptime of AI fabrics. The adoption of these open specifications is expected to reduce complexity and foster greater interoperability within the industry, benefiting all stakeholders involved in deploying and maintaining AI infrastructure.
What's Next?
The approval of these new specifications by the OCP Community marks a significant step towards industry-wide adoption of standardized optical telemetry. The next phase will involve the ecosystem building upon these specifications, with the OCP Foundation anticipating momentum in their implementation. The Enhanced Telemetry Optics Product Software Specification and the Transparent Remote Optical Transceiver Messaging with ePPS Design Specification are designed to provide a robust framework for optical transceivers and hosts to exchange link-health information. This will enable more intelligent diagnostic endpoints, giving operators better tools to address optical link flaps. The OCP Academy is also offering free courses on Optical Connectivity and Reliability for AI Infrastructure, which will help educate the industry on these emerging optical technologies and enhanced telemetry optics, further driving their integration and use.
Beyond the Headlines
The standardization efforts led by Credo and Oracle within the OCP Community have deeper implications for the future of AI and data center operations. By transforming optical transceivers into 'intelligent diagnostic endpoints,' the industry is moving towards a more proactive and automated approach to network management. This shift is vital as AI fabrics become increasingly complex and distributed, making manual troubleshooting impractical and costly. The emphasis on open, interoperable standards not only promotes competition and innovation but also reduces vendor lock-in, allowing for more flexible and scalable infrastructure deployments. This development reflects a broader industry trend towards collaborative efforts to solve complex technical challenges, ensuring that the foundational connectivity layers can keep pace with the exponential growth and demands of artificial intelligence.
