Leaf switches are popular devices used in data centers to aggregate traffic from server nodes and then connect to the core of the network, consisting of spine switches. In this model, there are only two tiers of switches between the servers and the core network. This popular data center design architecture is known as “leaf-spine” and was popularized as data centers grew in scale.
The leaf switch evolved in modern data centers as servers multiplied and network designs with more switching tiers – including perhaps an additional aggregation layer between the server access switches and the core – became less efficient as more racks of servers were added. This is especially the case in hyperscale data centers, where hundreds or thousands of servers might be connected to a network.
Where Did Leaf Switches Come From?
The terms “leaf” and “tree” relate to the description of a network. A tree network might have multiple tiers of switches, stacked upon each other, each getting larger in capacity. See the diagram below.
A leaf switch connects directly to the core – like a leaf would connect to a branch of a tree. With the leaf-spine design, every spine switch can connect to any leaf switch, connecting the leaf switches directly into the core data switching fabric of a network. See the diagram below.
Advantages Leaf Switches
The leaf-spine design has the advantage of placing each leaf switch just one hop away from another. There is no need to jump up and down the “tree” design. Latency is improved and bottlenecks are minimized. This system is equipped to handle east-west traffic from server to server, but it can also accommodate the north-south type, including emails and web content. This approach uses as few switches as possible to create a large, non-blocking fabric.
Google is widely credited with popularizing the leaf-spine design with its Clos architecture, which enabled the company to quickly add the new leaf and spine switches to accommodate data center growth, without re-architecting the network. This has become known also a “scale-out” architecture because new leaf-and spine combination can be added incrementally with new servers.
Leaf-spine architectures have many other advantages in addition to creating a more robust switching fabric with fewer tiers of switches. They can simplify cabling, for example, but reducing the cabling from the access switches on the server racks to the core. And with higher speeds being introduced for leaf and spine switches, these cables are typically high-capacity optical cables which are easier to manage than copper cables.
The expansion of the leaf-spine architecture, as well as increased gigabit Ethernet speeds, has improved the efficiency and capacity of data centers while at the same time making them easier to scale through the simple addition of more switching capacity.