Data Center Networking: The Spine-Leaf Fabric Inside the Data Center
Data center networking is the switching fabric that connects servers, storage and GPUs inside a data center. The modern design is spine-leaf: a flat, two-layer fabric where every access (leaf) switch connects to every spine switch and servers attach only to leaves, so any server is the same short distance from any other. Overlays such as VXLAN EVPN add segmentation and multi-tenancy on top.
This replaced the older three-tier model (core, aggregation, access) for a simple reason: traffic changed direction. Virtualisation, distributed applications and AI made east-west traffic, server to server and GPU to GPU, dominant, and three-tier networks were built for north-south traffic in and out of the data center. A fabric optimised for the wrong direction becomes the bottleneck that no faster server or storage array can fix.
Three-Tier or Spine-Leaf?
The architecture decides how the network performs as it scales. The table below sets out the difference, including the dedicated fabric AI now demands.
| Architecture | How it is built | Traffic it favours | Best for |
|---|---|---|---|
| Three-tier | Core, aggregation and access layers | North-south (in and out) | Legacy and smaller data centers |
| Spine-leaf | Leaf switches meshed to spine switches | East-west (server to server) | Virtualised, cloud and private cloud |
| AI back-end fabric | Dedicated lossless fabric between GPUs | GPU to GPU at 400G and beyond | AI training and inference clusters |
The Platforms: Cisco, Dell and HPE
Proactive builds data center fabrics on three platforms, matched to the operating model and the existing estate.
| Vendor | Switching | Fabric and automation |
|---|---|---|
| Cisco | Nexus 9000 (9300, 9500) | Cisco ACI with APIC, or NX-OS VXLAN EVPN managed via Nexus Dashboard |
| Dell | PowerSwitch (S and Z series) | SmartFabric Services with VXLAN EVPN, on OS10 |
| HPE | Aruba CX switching | Fabric with VSX resiliency and VXLAN EVPN |
The choice is rarely just the switch. It is the fabric and automation model, Cisco ACI's policy-driven approach, Dell SmartFabric's automation, or a standards-based VXLAN EVPN design, that shapes how the network is run day to day. Proactive designs to the model your team can operate, not the one that looks best in a demo.
Networking for AI: the Back-End Fabric
AI training spreads a single job across many GPUs that exchange data constantly, so the network between them has to be lossless and very low latency, or expensive GPUs stall waiting on it. This back-end fabric, typically RoCEv2 over lossless Ethernet with PFC and ECN congestion control at 400G or 800G and increasingly aligned to the emerging Ultra Ethernet Consortium (UEC) standard, is separate from the general data center network and designed specifically so GPUs are never the thing that waits. It is the networking half of an AI cluster, and it ties directly into AI infrastructure design.
Why Data Center Networking? Why It Matters Now
- Built for east-west: a spine-leaf fabric optimised for the server-to-server and GPU-to-GPU traffic that now dominates.
- Predictable at scale: consistent latency and easy growth by adding spines or leaves, not re-architecting.
- Segmented and multi-tenant: VXLAN EVPN overlays that isolate traffic and tenants on shared hardware.
- AI-ready: a lossless back-end fabric for GPU clusters, designed alongside the compute and storage.
- Automated and observable: policy and automation through ACI, SmartFabric or VXLAN EVPN, with telemetry built in.
- Designed to operate: a fabric and management model your team can actually run, with one accountable partner.
The data center network is the layer that decides whether everything else performs. A fast server and an all-flash array still crawl if the fabric between them is congested or built for the wrong traffic pattern. As workloads went virtual, distributed and now GPU-bound, the network became the constraint most often blamed on something else. Getting the fabric right is what lets the rest of the investment pay off.
Proactive Data Systems designs data center fabrics around the traffic they actually carry. As a Cisco Preferred Cloud and AI Partner, we build on Cisco Nexus, Dell PowerSwitch and HPE Aruba CX, with the overlay, automation and, where needed, the lossless AI back-end fabric, designed to be operated by your team and to scale without a rebuild.
Inside the Data Center, and Connecting It to the World
This page is about the fabric inside the data center: the spine-leaf network connecting servers, storage and GPUs. Connecting data centers to each other, to branch and campus sites, and to the cloud is a related but distinct discipline, covered by our Networks practice through SD-WAN, secure networking and data center interconnect.
The two meet at the data center edge and are designed together, so traffic flows cleanly from a GPU in the rack to a user across the country. Proactive designs both, which is why the hand-off between the internal fabric and the wider network is planned rather than improvised.
Data Center Networking Across India: Why the Workload Decides the Fabric
Indian enterprises arrive at a fabric refresh from different places. A GCC standing up a new private cloud is a different problem from a bank modernising a three-tier network it has run for a decade, or a manufacturer adding its first GPU cluster and discovering the existing network cannot feed it.
Workload mix, the size of the existing estate, AI plans and budget all shape the right fabric here rather than on a datasheet. Proactive has designed and built data center networks across manufacturing, BFSI, healthcare, IT and ITeS and GCC environments in Delhi, Mumbai, Bengaluru, Pune and Hyderabad, sizing each fabric to the traffic it carries and migrating from three-tier without a flag-day outage.
Proactive Data Systems: The Partner That Designs, Builds, and Runs the Fabric
Buying switches is easy. Designing a fabric that performs as it scales, migrating to it without downtime, and giving your team a model they can operate is the part that rewards experience.
Proactive brings over three decades of enterprise infrastructure delivery, certified engineers and an ISO 9001:2015 quality system. As a Cisco Preferred Cloud and AI Partner, we design and build data center fabrics on Cisco Nexus, Dell PowerSwitch and HPE Aruba CX, with ACI, SmartFabric or VXLAN EVPN and a lossless AI back-end where it is needed.
The data center fabric ties the rest of the stack together. It works alongside Compute Solutions, Storage, AI Infrastructure, Converged and Hyperconverged Infrastructure, and Data Protection and Cyber Recovery, and connects out through our Networks practice.
From design and platform selection through build, migration and ongoing support, backed by a 24/7 service desk, Proactive builds the fabric that lets compute, storage and AI perform at the speed you paid for.
