Composable Memory Fabric Architecture Enabling Scalable Memory Management for AI and General

Hongjian Fan Electrical Engineer - Seagate, Anjaneya "Reddy" Chagam Senior Principal Engineer - Intel

The rising demands of workloads such as in-memory databases, caching, and AI are driving the need for scalable memory expansion, pooling, and sharing solutions. Cache-coherent interconnects like Compute Express Link (CXL) provide mechanisms to dynamically provision memory resources tailored to workload requirements. The OCP CMS Memory Fabric Orchestration Workstream is dedicated to developing a composable memory orchestration architecture by collaborating with standards organizations, enabling open and interoperable solutions, and supporting ecosystem vendor reference implementations to accelerate broad market adoption. This technical talk presents an in-depth overview of the memory fabric architecture, which leverages CXL and DMTF interfaces to empower data center orchestration tools, such as Kubernetes, to efficiently manage memory resources across the data center. The talk will also highlight future focus areas for the CMS Memory Fabric Orchestration Workstream.