Because of the way traffic flows in modern data centers, the traditionally deployed three-layer hierarchical model...
that dominated network architectures in the past may no longer be ideal. Thanks to virtualization technologies, server administrators can now effortlessly move compute resources from one part of the data center to the next with a few clicks of the mouse. Yet, the underlying network remains static in nature. Therefore, if a dramatic shift in data flows were to occur, it can create performance issues that are not easily resolved. One solution to this dilemma is to move toward a leaf-spine network architecture.
But just what is leaf-spine? And, more importantly -- how would I know if I would benefit from one? In this article, we're going to first look at why leaf-spine architecture might be a preferred option, given today's data center requirements. Next, we'll move on to point out cues that should trigger you to act and implement a leaf-spine fabric.
Leaf-spine network architecture overview
Even though leaf-spine technologies are being marketed as a groundbreaking, new concept, the overall architecture dates to the 1950s. Charles Clos, an engineer at Bell Labs, is credited with coming up with a mesh design for networks that required multiple, nonblocking switching paths between locations. This multipath design provided added capacity by enabling ports that would previously have been blocked. Original uses for the Clos network at the time were geared toward adding capacity into the public switched telephone network. In the 1990s, Clos' concepts were adopted and integrated into switch cross-bar backplanes to significantly increase interport bandwidth within a single switch.
And today, the Clos design is expanding into the LAN to offer a mesh of nonblocking interconnections inside the data center. The idea is to provide a boost in east-west traffic capacity and more predictable communication times between data center resources. This is what is being referred to as the leaf-spine architecture touted by most network vendor marketing departments today.
Identifying the need for a leaf-spine architecture
So, now that we have a basic understanding of a leaf-spine architecture, we can move on to discuss happenings within your data center that point to the possibility of benefiting from this type of design. These identification cues can be broken down into three main categories: data flows, latency and management tools.
It used to be that the vast majority of data center data flowed between the client and server. This type of data flow is referred to as north-south data flows. But, thanks to virtualization technologies, server resources are now distributed in different physical areas of a data center. This creates the need for ultra-fast transport in an east-west manner. As your data center becomes increasingly virtualized, you'll likely see a matching increase in east-west traffic proliferation. And before you know it, bottlenecks form as a result of single-path connections between one portion of the data center and the next. Therefore, it's important to monitor east-west link utilization closely. And when it reaches a predetermined threshold, it's probably time to consider a change to a leaf-spine architecture.
Another similarly related -- yet more difficult to identify -- trait deals with the sensitivity of applications and resources as they relate to network latency. Again, because server resources are distributed -- and because legacy data centers use a single-path tree architecture -- the physical location of compute and storage components that comprise a virtualized server becomes critical. If deployed incorrectly, resources could be several network hops away. This increases latency, which ultimately affects the overall usability of many applications.
A leaf-spine architecture, on the other hand, flattens the data center because of the mesh design of nonblocking uplinks between leaf and spine switches. Ultimately, it reduces the number of network hops between any two given resources. The trick is to be able to identify latency issues that are network-related, as opposed to application-related. Network performance monitoring tools are a great resource to assist in this identification. When network latency becomes a serious issue, it's a sign that your current network architecture is failing to provide what you need.
One final indication you may be ready for a leaf-spine network architecture: You find yourself seeking advanced tools to deal with the overall management of a data center network. For example, if your network administrators are struggling to keep up with application deployment demands, then you may require the use of a leaf-spine network that provides automated deployment capabilities.
Another example is if you need a data center that can be segmented in a Multi-tenancy fashion. If this is the case, Multi-tenant provisioning tools that come along today's leaf-spine deployments are going to be a great fit. Finally, if your applications and data require end-to-end visibility to manage complex traffic flows -- or for implementing advanced security tools -- the proper way to gain these abilities is to look into a data center architecture that includes leaf-spine fabrics and software-defined networking technologies.
Making a decision on leaf-spine
If you think a leaf-spine network architecture is only for enormous data centers with highly advanced networking needs, you're sorely mistaken. There are plenty of reasons why data centers, both small and large, might be ready to move to the next-generation design. You simply need to know how to identify those reasons. If your data center is suffering from bottlenecks, slow application performance or requires more advanced network management tools, 2017 is the year to consider the wide range of leaf-spine architecture options available.
The case for a leaf-spine topology
Five ways to improve your data center
Leaf-spine designs meet data center demand