The leaf-spine network topology, also known as the spine and leaf network topology, was created to more efficiently accommodate modern data center traffic, which has different characteristics as compared to traditional enterprise network client-server traffic. Data center network traffic tends to be more “east-west” in nature – traveling between servers, or between servers and storage systems that are largely side by side. Generally only a limited amount of “north-south” traffic travels between a server and a client located outside the data center.
Traditional client-server networks have three tiers. Access layer Ethernet switches are closest to end users and their various devices. They feed into aggregation tier switches or routers (also known as the distribution layer), which then feed into “core” routers at the top layer.
These networks are configured such that multiple, redundant pathways exist between any two endpoints, so traffic can route around any network failures. To prevent excessive bandwidth consumption from network loops and broadcast packets, the networks typically employ the Spanning Tree Protocol (STP). With STP, traffic always has two potential routes to take, but only one is active at any given time. If the primary path fails, STP will route traffic over the backup path, but only until the primary is back up.
While STP generally works well in three-tier client-server networks, it can result in backups on links that experience heavy loads, which are common in data centers.
The leaf-spine topology is designed to address that limitation by reducing the number of “hops” in the network design. The leaf-spine architecture (spine and leaf architecture) comprises just two layers: leaf and spine switches.
Leaf switches are analogous to the access layer, as they connect end devices in the data center, such as servers, firewalls, edge routers and load balancers. To ensure redundancy, each leaf switch connects to multiple spine layer switches, perhaps even to all of them.
For example, at the top of each data center server rack may sit two leaf switches, known as Top of Rack (ToR) switches. Each server in the rack connects to both of the ToR switches, to ensure redundancy.
image of leaf-spine networks
Each ToR switch also connects to multiple spine layer switches. In some instances, each leaf switch may connect to every spine switch. That means each leaf switch – and each device connected to those switches – is just a single hop away from any other device in the data center. That makes the leaf-spine design a good fit for the “east-west” nature of data center switch traffic, dominated by traffic flows between servers – and delivers a significant performance improvement.
Also, because every leaf switch can connect to every spine switch, the spine switches don’t have to connect to each other. Rather, all their ports can be used to support leaf switches, using either Layer 2 switching or Layer 3 routing.
What’s more, the protocols used with leaf-spine topologies allow for traffic to flow across all available routes, not just one at a time as with STP.
Leaf-spine networks provide a number of benefits as compared to traditional three-tier client-server network designs, including:
- Increased redundancy, because each access switch connects to multiple (if not all) spine switches, along with the use of protocols such as Transparent Interconnection of Lots of Links (TRILL) and Shortest Path Bridging (SPB), that allow traffic to flow across multiple available routes.
- Improved performance, from the ability to use numerous network paths at once, as compared to only one at a time with STP, and from having only a single hop between any two end points.
- Scalability, because adding switches to a leaf-spine network provides additional traffic routes, thus increasing scalability.
- Leaf and spine switches can also be less expensive because they can be built using commodity, fixed configuration switches rather than expensive modular chassis-based switches, which are often required to deliver the port density that three-tier networks require. For the same reason, the leaf-spine design is a good fit for white box networking.
While it’s true that the leaf-spine network architecture was developed with data centers in mind, some vendors are now extending the concept – and its various benefits – to the larger enterprise network.
The white box networking vendor Pica8, for example, has a solution called PicaPilot that enables an enterprise-wide leaf-spine network. When used with the Pica8 PICOS network operating system, the result is a network that has all the features and functions of a traditional three-tier design – including Layer 2/Layer 3 switching and routing and software-defined networking (SDN) – but with a much simpler, flatter architecture that’s easier to manage.
PicaPilot enables hundreds of leaf and spine switches to appear as a single logical switch, with a single IP address. That means a network administrator can literally manage hundreds of switches as if they were one – dramatically lowering operations costs. PICOS also runs on a range of white box switches, making it even more cost-effective.
Pica8’s PicaPilot is application software that runs alongside the company’s PICOS NOS on every switch and gives customers a choice of four pre-defined workflow templates – one for chassis switch replacement and three for switch stacks – that cover the vast majority of enterprise deployment cases, allowing customers to provision, manage and gain more visibility into their new branch office and campus infrastructure. This software is explicitly designed to help enterprise customers of all sizes who need to both reduce OpEx and upgrade their networks to a modern two-tier architecture, moving away from archaic and costly switch-stacks and chassis to an easily deployed leaf-and-spine architecture for the first time.
When the concurrent demands of ever-evolving network operations, big data initiatives, and client virtualization are overlaid on top of the budget constraints imposed by expensive, traditional 3-tier network infrastructure, today’s enterprise access networks are simply unable to keep up with the frantic pace of organically changing business data requirements. In the data center itself, 2-tier leaf-and-spine deployments have become one of the most common network architectures used by the industry. This proven new design offers exactly the kind of flexibility and scalability benefits that could be of tremendous value to the entire enterprise network – if the complexity of deployment and management that exists today can somehow be extracted from the architecture.
To accommodate the ever-increasing need for bandwidth at the access layer in a wiring closet, Pica8’s PicaPilot utilizes the well-known “leaf-spine” Clos architecture model to scale out the network horizontally. In this case, the “spine” switches are the equivalent of controllers and “leaf” switches form the access layer that delivers networking connection points for servers. Recognizing the requirement to change today’s monolithic way of deploying management software and automation, Pica8 is now the first in the open networking industry to deliver both orchestration and zero-touch provisioning (ZTP) via its PicaPilot software.
Pica8’s PicaPilot is built on an innovative network-wide leaf-spine architecture and is the first switch management tool for white box networking that offers easy provisioning, configuration, image management, troubleshooting, visibility, and security.