Network Rack Functional Architecture

Network rack architecture organizes and interconnects compute, storage, and networking resources within racks to optimize performance, scalability, and manageability in modern data centers.

Traditional Networking Racks

A networking rack is a metal frame designed to house rack-mountable devices such as routers, switches, patch panels, and UPS units, providing a structured and secure environment for network equipment. Standard racks are typically 19 inches wide, with heights ranging from 48 to 83 inches, and vertical spacing measured in rack units (U), where 1U equals 1. 75 inches.

Accessories like cable organizers, sliding shelves, and vented panels enhance functionality, improve airflow, and simplify maintenance.

Rack-

Scale Architectures (RSAs)Modern data centers increasingly adopt rack-scale architectures, which integrate compute, storage, and networking into cohesive units called rack nodes or pods. Each pod contains application storage, compute resources, and top-of-rack (ToR) networking, often connected via high-speed fabrics such as 25–400 GbE. This design reduces latency by keeping most application traffic within the pod and allows pods to scale out efficiently across rows using leaf/spine networks.

Disaggregated and Composable Racks

Disaggregated racks separate compute, storage, memory, and networking into independent modules that can be dynamically composed into logical systems based on workload requirements. Intel® Rack Scale Design (RSD) exemplifies this approach, enabling software-defined composition of resources, flexible management, and efficient utilization of hardware. Each module typically has dual network connections: a high-bandwidth data link and a management link, often connected to ToR switches and orchestrated via open APIs like Redfish.

Power-

Efficient Intra-

Rack NetworkingHigh-density disaggregated racks require efficient intra-rack networks. Solutions like Shoal use fast circuit switches to interconnect numerous resource nodes with minimal power consumption and nanosecond reconfiguration. This approach eliminates in-network contention, provides bounded throughput, and achieves high performance while reducing energy usage compared to traditional server-centric architectures.

Network Topology Considerations

Two common network designs in racks are: Top-of-Rack (ToR): Switches are installed at the top of each rack, connecting directly to servers within the same rack. This simplifies maintenance, isolates faults, and reduces cable clutter.

End-of-

Row (EoR): Aggregation switches are placed at the end of each row, connecting all servers in the row. This reduces switch costs but requires extensive cabling and can impact multiple racks if a switch fails.

Summary

Network rack architecture has evolved from traditional 19-inch racks housing fixed servers and switches to dynamic, disaggregated, and composable designs that optimize resource utilization, reduce latency, and improve scalability. Modern approaches leverage high-speed fabrics, software-defined orchestration, and power-efficient intra-rack networking to meet the demands of hyperscale and enterprise data centers.