It’s time to consider leaf-spine network architecture

With the changes in traffic flows used in modern data centers, the three-tier network architecture that dominated in the past is no longer an ideal solution. Thanks to the development of virtualization technology, server administrators can now easily migrate computing resources from one location in the data center to another with just a click of the mouse. However, the underlying network is still static in nature. Therefore, if the data flow changes significantly, it will still cause some performance issues that are not easy to handle. One solution to this problem is to switch to a leaf-spine network architecture.

But what is leaf-spine? And, more importantly - how do I know that this architecture will benefit my network? In this article, we will first explain why leaf-spine architecture should be the first choice for modern data centers. Then, we will introduce some key reasons that motivate people to use and implement a leaf-spine architecture.

Leaf-Spine Network Architecture Overview

Although leaf-spine technology is still considered a new technology and concept, this architecture was proposed as early as the 1950s. One of the important contributions of Charles Clos, an engineer at Bell Labs at the time, was the mesh network design, which required the use of multiple non-blocking switching paths between two locations. This multi-path design can enable previously blocked ports to add greater capacity. Clos networks were designed to add greater capacity to the public switched telephone network. In the 1990s, the concept of Clos was adjusted and integrated into the switch cross-rack backplane, greatly improving the bandwidth between ports in a single switch.

Now, Clos design has been extended to LAN, thus forming a mesh-like non-blocking interconnection structure within the data center. This concept can greatly increase the east-west traffic capacity and achieve more controlled communication between data center resources. This is what most network vendors' sales departments now refer to as leaf-spine architecture.

Leaf-spine networks use a mesh of non-blocking connections to improve traffic flow

Determine the need for a leaf-spine architecture

Now that we understand the basic concepts of leaf-spine architecture, we will discuss the changes and benefits that this design can bring to the data center. These landmark clues can be divided into three categories: data flow, latency, and management tools.

In the past, there were all kinds of data center data flows between clients and servers. This data flow is called north-south (vertical) data flow. However, with the advent of virtualization technology, server resources are now distributed across different physical locations in the data center. This has led to the need for ultra-high-speed east-west (horizontal) data transfer. As data centers become more virtualized, east-west traffic will continue to grow. This can create a bottleneck in the connection path between two locations in the data center. Therefore, it is important to closely monitor the usage of the east-west link. When it reaches a predetermined critical value, it may be time to consider switching to a leaf-spine architecture.

Another similar, but harder to spot clue is dealing with intensive applications and resources that can cause network latency. Again, because server resources are distributed -- and legacy data centers use a single-path tree structure -- the physical location of the compute and storage components that make up a virtual server becomes critical. If not properly deployed, these resources can be separated by several network segments. This increases latency and ultimately affects the overall availability of many applications.

On the other hand, the leaf-spine architecture flattens the data center because the non-blocking uplinks between the leaf-spine switches are meshed. Ultimately, it reduces the number of network hops between any two resources. The hardest part is to determine if the latency is caused by the network, not the application. Network performance monitoring tools are a great resource to help determine the source of the problem. When network latency becomes severe, it means that the current network architecture is no longer adequate.

The most obvious sign that you should adopt a leaf-spine network architecture is that you need to find some advanced tools to manage the entire data center network. For example, if the network administrator cannot cope with the application deployment requirements, then you may need to use a leaf-spine network to provide automatic deployment capabilities.

Another example is when you need to support multi-tenant segmentation in the data center. If this is the case, then the multi-tenant allocation tools that come with modern leaf-spine deployments are a good choice. Finally, if the applications and data require end-to-end visibility to manage complex traffic flows - or implement advanced security tools, then the right way to get these capabilities is to look for a data center architecture that includes leaf-spine and software-defined networking technologies.

Deciding on leaf-spine architecture

Don't think that leaf-spine network architectures are only for large data centers with advanced networking needs. This is a misconception. No matter how big or small, there are many reasons to move to a next-generation design. It's the administrator's job to understand how to identify those reasons. If the data center is experiencing network bottlenecks, application performance degradation, or requires more advanced network management tools, then 2017 is the time to consider the various leaf-spine architectures.