Phils Photography - stock.adobe.

Build up your knowledge of leaf-spine network technology

As organizations deal with challenges of single tree path networks, many admins are considering leaf-spine architectures. Learn the ways its topology helps latency and data flow.

Leaf-spine technology is a network topology that moves away from the more traditional network architectures found in older data centers that have a single tree path design. Its distinguishing factor is that its switches support east-west network traffic.

You can set up either a two- or three-layer leaf-spine architecture in your data center. The switches mesh into the spine and connect to every switch in the network fabric. Because this model provides a consistent number of devices to disperse and transfer data, the topology reduces latency and bottlenecks in the data center.

Current providers for leaf-spine networks include Cisco, Mellanox and Arista Networks.

How can a leaf-spine topology help network traffic?

Most data center networks support north-south traffic that goes between client and server or somewhere outside the data center.

However, new types of data center technology and infrastructure can experience bottlenecks with a north-south network traffic model. Leaf-spine networks take a more horizontal approach to network design and include one or two tiers instead of three.

With leaf-spine architecture, each switch gathers and consolidates user traffic from clients and then feeds it through a central network line to the server. This model helps reduce the number of required switches, as well as network maintenance and deployment costs.

This setup consolidates and characterizes network traffic into packets and moves them between servers. Because devices don't need to wait for open server connections, this setup can reduce overall latency in the data center.

Additional benefits of leaf-spine network designs include equidistant switches for east-west traffic that improve reliability, support for all types of interconnection links to help scale networks and compatibility for fixed configuration switches.

Leaf-spine network topology
As a more horizontal network topology, a leaf-spine architecture often uses two or three layers.

Your organization may also consider fabric extenders as one way to address network issues. Though the technology does give you the ability to build out network connectivity to edge devices, fabric extenders must run traffic through a spine layer and be paired to a virtual private cloud to run.

Leaf-spine topology, on the other hand, has local packet switching and multiple spines to improve network resiliency, are entirely independent systems and reduce the possibility of single-point failure.

What are the driving factors for leaf-spine network adoption?

Your organization may naturally shift to leaf-spine architecture as it upgrades its infrastructure, or it may decide to have you evaluate market options before your team takes on an entire network hardware upgrade. Data flows, latency and management tools are driving factors in the move to a leaf-spine network design.

Data flows, latency and management tools are driving factors in the move to a leaf-spine network design.

Single-path connections between servers and clients are the de facto connection in most data centers. Though with more virtualized infrastructure, you might see more east-west traffic as server resources are more equally distributed across the data center.

When you monitor your network traffic, pay attention to how much east-west activity your data center is producing. If you see a growing amount of east-west traffic -- or a regular level of activity -- it makes the most sense to use leaf-spine topology to support this traffic.

Leaf-spine networks are also suitable for any hyper-converged infrastructure, so if your organization decides to deploy HCI, you should decide if your network needs an upgrade.

Latency is another indicator that your organization may benefit from leaf-spine networks, because the design reduces the number of required pings across the network. If you find that your network has high latency levels or is regularly slow, then centralizing traffic through leaf-spine is one way to address concerns.

Network monitoring is another factor for leaf-spine implementation. If you want more advanced network management and monitoring tools to properly evaluate your network, this is where a leaf-spine architecture can help. Once you have a leaf-spine setup, you and your team can implement automated deployment capabilities for more streamlined management.

Additionally, if your data center requires multi-tenant segmentation, provisioning tools from leaf-spine setups are a natural step. This functionality, along with end-to-end network visibility is available through most leaf-spine and software-defined networking offerings.

Are there drawbacks to leaf-spine networks?

Even with all the benefits of leaf-spine architectures, you should still consider the drawbacks before deciding to modify your data center. Though it brings centralized network traffic processing, there are still a high number of hardware requirements and traffic ratios to address.

A large drawback is the required scale for a leaf-spine topology, which directly correlates to the number of physical hosts you want to support. The more hosts you must support, the larger the spine. In general, a spine can only extend so far before its switches run out of ports, cannot support interconnection between switches or it reaches an oversubscription rate that is too high.

Leaf-spine oversubscription rate
Leaf-spine networks require the right scale and subscription rates.

Leaf-spine architectures have a lot of cabling requirements. The more spine switches you have, the more cables your setup requires. The number of necessary cables is also dependent on the number of spine switches you have; the wider the spine, the more interconnect switches you need.

Distance of interconnects can drive up costs. Depending on how far interconnecting switches are from each other, they may require optical modules, which increases the overall deployment cost.

Dig Deeper on Data center hardware and strategy

Cloud Computing
and ESG