Ethernet switching designs for data centers have changed significantly over the last decade and now include a spine switch as part of that data center architecture. In recent years, an architecture known as leaf and spine, or distributed core, has emerged as a leading design for data centers. This design requires spine specially designed spine switches in order to work.
A traditional network fabric model included three tiers of switches: Core, aggregate, and access. This model was sufficient for basic network set-ups and vertical traffic. However, with the growth of networks in data centers created the need for faster and more efficient lateral – or “East-West” traffic. In addition, many data centers employ Spanning Tree Protocol, which prevents network loops, but can block up to 50% of switch ports and limit broadband use. This is where the leaf-and-spine architecture became popular.
Spine Switch: Core piece of a Networking Fabric
All spine switches can handle Layer 3 (L3) with high port density, which allows for scalability. In a software-defined network (SDN), the spine switch is directly connected to a network control system with a virtual Layer 2 switch on top of the leaf-spine system. This controller is not a part of the data path but remains the central network engine, also known as an SDN Controller.
Every L3 port on a spine switch is dedicated to connecting to underlying (L2) leaf switches. No servers, access points or firewalls can be connected or found on a spine switch, they only serve to connect to the leaf switches. One method of connecting is using equal cost multipathing (ECMP). In this architecture, every leaf switch is connected to every spine switch. As such, the amount of uplink outlets on a leaf switch determines the number of spine switches you can have in your network. This model allows for every connection has the exact same distance to travel, there are no discrepancies in the fabric.
Benefits of a Spine-Leaf System
Some of the benefits of this system involve some predictability with latency. With a leaf-spine architecture, you always know the amount of stops that each packet of information must travel. In the old architecture, payload from one server may have to travel through several switches before it can communicate with a different server on the same network. Thus, there is consistency with the traffic and distances it must cover, also leading to fewer bottlenecks between access-layer switches. In addition, the removal of the STP in favor of this Level 3 routing makes for a far more stable environment.
One other benefit of this structure is scalability. When there is too much traffic for one active link, adding additional spines increases the interlayer bandwidth and reduces the amount of “traffic jams” in the network. Adding new spines or leaves into a system also will not disrupt the L2 protocols.