Is your data centre network ready for what comes next?

‍

Applications and infrastructure have changed significantly and at a very fast pace over the last decade, but in many organisations the data centre network has evolved more slowly.

Virtualisation, hyperconverged infrastructure, container platforms and hybrid cloud are placing more demands on the data centre network. Customers are increasingly talking to us about support for greater east-west traffic flows, stronger segmentation and more dynamic provisioning capabilities.

So when organisations undertake a data centre refresh, the question shouldn't just be what switches do we need? It should also be is the underlying network architecture still right for what we need it to do?

The challenge with traditional designs.

Traditional data centre networks have usually relied on extending Layer 2 VLANs across the DC switching environment. Spanning Tree Protocol prevents forwarding loops, while first-hop redundancy protocols provide resilient default gateways for the connected workloads.

These designs do remain valid and continue to support many production environments. However, as the number of VLANs, security zones, racks and data centre locations increases, they have become difficult to scale and operate.

Extending Layer 2 domains within or between DCs also increases the potential failure domain. Changes to VLAN trunks, spanning-tree topology or gateway placement can have effects well beyond the immediate area of change. While traditional data centre architectures can still support modern requirements, doing so often introduces additional complexity.

How EVPN-VXLAN changes the architecture.

EVPN-VXLAN separates the physical data centre network fabric from the logical networks deployed across it.

The physical underlay is normally built as a routed leaf-spine fabric. Every leaf switch connects to each spine, providing multiple active paths across the network. Equal-cost multipath routing allows traffic to use all available links rather than leaving redundant, high capacity paths idle.

VXLAN provides the overlay. Layer 2 and Layer 3 services are represented by VXLAN Network Identifiers, or VNIs, and transported between VXLAN Tunnel Endpoints located on the leaf switches. This allows logical networks to be deployed independently of the physical topology.

BGP EVPN provides the control plane, advertising endpoint and network reachability information between the leaf switches. This reduces reliance on traditional Layer 2 extension and data-plane learning while still supporting Layer 2 connectivity across the routed fabric where required.

Distributed anycast gateways provide a consistent default gateway across the EVPN-VXLAN fabric. Each leaf can act as the gateway for its locally connected workloads, helping to support workload mobility and more efficient traffic forwarding.

Why customers are interested.

• More Predictable Scale: Additional racks and leaf switches can be introduced into the DC using the same repeatable design. The routed underlay provides IP reachability, while the overlay determines which logical services are available at each leaf. This gives organisations a more consistent way to scale out the networking fabric in the DC as needed.
• Stronger Segmentation: EVPN-VXLAN supports multi-tenancy through the use of Layer 2 and Layer 3 VNIs combined with VRFs. This enables application environments, security zones and tenant networks to remain logically separated while sharing a common physical leaf-spine fabric.
• Better Use of Available Bandwidth: A leaf-spine topology uses Layer 3 routing and ECMP to provide multiple active paths through the fabric. This differs from traditional spanning-tree designs, where redundant links may be blocked to prevent loops. The result of this is more available DC capacity to support east-west connectivity and fast recovery when a link or switch in the fabric fails.
• Automation and Orchestration: Modern data centre platforms increasingly support APIs and integrate with controllers and orchestration software. Network segments, VRFs and associated policies can then be delivered through a single orchestration platform rather than configured individually on each switch as would have been completed historically. This is a very relevant and significant benefit where compute, storage and networking need to be provisioned together.
• Supporting Modern Traffic Pattern: Modern applications generate more east-west traffic within the data centre than traditional client-server applications. Virtualisation, container platforms, analytics applications and AI workloads place bigger demands on the Datacentre network. The leaf-spine architecture with EVPN-VXLAN provides predictable low-latency connectivity and multiple active forwarding paths, making it very well suited to supporting these modern traffic patterns.

EVPN-VXLAN isn't just for large datacentres.

EVPN-VXLAN was initially associated with large data centre environments, but that is no longer the case. Improved segmentation, active-active forwarding, consistent configuration and a network fabric that supports greater automation may also be relevant to organisations operating smaller data centre environments.

That doesn't mean every data centre refresh needs EVPN-VXLAN. Adding an overlay and BGP EVPN also introduces new design and operational considerations, and infrastructure teams need the skills and tooling to operate the environment effectively. That is why the refresh decision should start with the requirement rather than the technology.

For organisations planning their next data centre investment, simply replacing existing switches with greater capacity and higher port counts may miss the bigger question.

Does the architecture itself still suit the applications you run today, and can it support where your infrastructure is heading next?

If the requirements increasingly involve distributed applications, greater east-west traffic, stronger segmentation and more automated provisioning, then the network architecture deserves as much attention as the hardware being refreshed.

‍