Why Networks Are Evolving Toward Leaf-Spine Architectures

Whether you refer to it as a “three-tier” or “tree” architecture, traditional enterprise networks are based on a strict hierarchical model that includes a core, distribution, and access layer. However, advancements in connectivity options combined with advanced software-defined network (SDN) technologies are causing network architects to reimagine how networks should be built. The name of the game is efficient data flows from end-to-end. If speed and latency are of concern, leaf-spine architectures are proving to be a better choice despite their higher cost to implement.

The primary benefit of a leaf-spine architecture is that it allows data flows to take shortcuts from where data is, to where it is going. Layer 3 data flows within traditional three-tier architectures can vary depending on where the source and destination devices reside on a network. In some cases, the data flows may only have to move up one level to the distribution layer to reach a destination. In others, the flows may have to reach all the way up to the core, then back down the stack. The three-tier architecture creates differences in speed and latency which can be problematic in modern enterprises where data flows are getting larger, and applications are becoming increasingly time-sensitive.

Data flows within a leaf-spine fabric, on the other hand, take the same number of hops on the network regardless of the source and destination. The primary reason for this is that a leaf-spine architecture is fully-meshed as opposed to three-tier models that are only partially meshed. While many may think that a fully-meshed architecture creates far too many physical interconnects to manage, large 25, 40- and 100-Gbps Ethernet links considerably reduce the number of physical ports required. This is due to the increase in data can flow across single links as opposed to multiple, aggregate connections.

(Source: Dabarti CGI/Shutterstock)

The key benefit to today's leaf-spine architectures is that each link within the full-mesh can be used and load balanced in a loop-free environment. We’re talking true active-active data flows here. This is in opposition to spanning tree protocol (STP) blocking ports at layer 2 of the OSI model -- or preferred dynamic routes placed in the routing table at layer 3. The organization of this type of equal-cost, multipath design inherent in leaf-spine fabrics, is at its best when a centralized network management platform such as SDN is used. The reason for this is that SDN allows for streamlined configuration, management, and re-routing of traffic when congestion or link failures occur. In other words, SDN makes a full-mesh topology that’s intelligently load balanced a relatively simple thing to configure and manage.

In today’s enterprise environments, leaf-spine architectures achieve the most observable benefits when deployed in the data center. The reason for this is that the efficiencies in data flows are most easily found in east-west data flows as opposed to north-south flows. East-west flows are server-to-server communication within the same data center. When using distributed server architectures where resources for a specific application or service are scattered across multiple servers, shortening the communication path between servers can significantly increase application and service performance.

Additionally, thanks to advancements in WAN and public/private cloud connectivity options, inter-data center communications using leaf-spine architectures is quickly becoming a reality. Building a leaf-spine fabric between data centers or clouds allows for similar speed, performance, and redundancy no matter if flows are intra- or inter-data center. Considering that multi-cloud is becoming a “must have” strategy within medium and large enterprises, it’s great to see that leaf-spine models are a viable option.