As workloads intensify from cloud services to AI/ML to containerised microservices, traffic inside data centres is no longer a simple matter of north–south (client-to-server) flow. Instead, vast volumes of traffic move laterally, east–west, between servers, storage arrays, compute clusters, and services. Traditional three-tier architectures (core → aggregation → access) start to show strain under such load.
DC-Clos architecture is a spine-leaf design that delivers non-blocking throughput, predictable latency and seamless scalability.
When powered by a disaggregated NOS like OcNOS, DC-Clos gives data centre and cloud operators a future-ready fabric that handles the demands of virtualisation, containerisation, AI and cloud-native workloads without compromising performance or flexibility.
Figure 1. High-level DC-Clos architecture
What Makes DC-Clos So Effective: Core Benefits with OcNOS
Non-blocking Performance and High Throughput
In a true non-blocking architecture, the aggregate uplink capacity from leaf switches to spine switches matches (or exceeds) the sum of all server-side ports. This eliminates oversubscription bottlenecks, meaning every server can communicate at full line rate simultaneously, without fear of congestion.
For AI/ML clusters, distributed storage, heavy east–west workloads or cloud-native microservices, this ensures consistent, high-speed connectivity across the data centre, critical for performance and reliability.
Low and Predictable Latency
Because DC-Clos maps to a two-hop path (leaf → spine → leaf) between any two endpoints, latency is consistent and minimal. That predictability matters enormously for latency-sensitive workloads like distributed databases, real-time analytics, GPU clusters, or storage-area network traffic.
Scalability and “Pay-as-You-Grow” Economics
Need more capacity? Just add more leaf or spine switches. With DC-Clos you don’t need disruptive “rip-and-replace” upgrades – you expand horizontally, modularly. That makes growth incremental, budget-friendly and operationally more manageable.
This modular expansion is ideal for evolving demands: new AI workloads, sudden spikes in demand, growing user bases or scaling services – all while preserving investment in existing infrastructure.
Overlay Flexibility for Multi-Tenant & Cloud-Native Workloads
On top of the physical underlay, overlay networking using standards such as EVPN‑VXLAN allows creation of virtual Layer 2 networks over the Layer 3 fabric. This means virtual machines, containers or microservices can be placed anywhere across the fabric, while retaining logical LAN connectivity.
Benefits include:
- Up to millions of logical networks (versus VLAN-limited) — enabling large-scale multi-tenant deployments.
- Mobility: services, VMs or containers can move freely without rearchitecting physical connectivity.
- Isolation: each tenant or application gets logical separation, even on shared physical infrastructure.
Deploying OcNOS-Based DC-Clos: What It Looks Like in Practice
Here’s a typical deployment flow using OcNOS:
- Build a leaf-spine underlay: leaf switches connect to servers, spine switches interconnect all leaves.
- Use Layer 3 routing underlay (e.g. eBGP or other standard routing protocols) for scalability and vendor interoperability.
- Overlay EVPN-VXLAN on top: leaf switches act as VXLAN Tunnel Endpoints (VTEPs), encapsulating and decapsulating traffic. This provides a virtual Layer 2 network across a Layer 3 fabric.
- Implement distributed gateway and routing logic (e.g. IRB — Integrated Routing and Bridging) at leaf or spine layer to enable efficient inter-subnet routing.
The result: a highly resilient, scalable, and flexible network capable of supporting cloud-native and legacy apps alike, with seamless VM/container mobility, multi-tenancy and simplified network operations.
Figure 2. OcNOS DC Clos roles and protocols
Why Cloud, AI and Modern Data Centres Need DC-Clos
Traditional three-tier or monolithic chassis-based network designs struggle to keep up with modern demands: unpredictable east–west traffic patterns, microservices proliferation, container churn, bursty AI workloads, storage-area fabric demands and multi-tenancy.
DC-Clos with OcNOS solves these problems:
- Burst-resilient fabric — data flows efficiently without congestion, even under high east–west load.
- Latency-sensitive support — two-hop paths and lossless transport support real-time, latency-dependent workloads (e.g. AI, high-performance computing, live data analytics).
- Operational agility — overlay networks, virtual fabrics, multi-tenant isolation and rapid provisioning.
- Cost-effective scale — modular expansion, whitebox hardware options, reduced vendor lock-in and lower total cost of ownership.
For organisations scaling cloud, hybrid-cloud, AI clusters, storage fabrics or multi-tenant services – DC-Clos gives the foundation they need, without overpaying for monolithic gear or sacrificing agility.
Figure 3. OcNOS Multi-DC Fabric
The IDS Advantage: Implementing Open, Scalable Fabrics in ANZ
As a partner and integrator offering open-networking solutions across Australia and New Zealand, IDS brings deep expertise in deploying DC-Clos fabrics powered by OcNOS.
We help clients:
- Design architecture: size leaf-spine fabric appropriately (port count, oversubscription, redundancy, scaling path)
- Deploy overlay networks: EVPN-VXLAN, tenant segmentation, VM/container mobility
- Integrate storage, compute, AI clusters, DCI links, multi-site interconnect – ensuring consistent, high-performance connectivity
- Provide ongoing support, lifecycle management, and vendor-agnostic flexibility to evolve infrastructure without lock-in
For businesses seeking to modernise data centres, support cloud-native workloads, or prepare for AI/ML demands – IDS combines the advantages of open-networking with local support, regional understanding and global best practices.
Build Today for Tomorrow’s Demands
As data centre workloads shift, from traditional hosting to cloud-native applications, microservices, AI, high-performance storage and multi-tenant cloud services, network fabrics must evolve accordingly.
DC-Clos architecture, underpinned by a flexible, open NOS like OcNOS, delivers a resilient, scalable, high-performance foundation that responds to modern traffic patterns, latency demands, multi-tenant complexity and rapid growth.
For organisations ready to build or upgrade their multiple data centre backbone, DC-Clos with OcNOS provides:
- Non-blocking throughput for predictable performance
- Low latency for latency-sensitive workloads
- Modular scalability with pay-as-you-grow economics
- Virtualisation-ready overlays for flexibility and multi-tenant services
- Reduced hardware lock-in and enhanced vendor choice
At IDS, we believe this architecture isn’t just ideal, it’s essential for future-ready data centres. If you’re considering a new build, an upgrade, or migrating to an open-networking, software-defined fabric, we’d be glad to assist with design, deployment and ongoing support.
Let’s build the flexible, scalable, resilient data centre fabric your organisation needs – ready for what comes next.
