Spine and Leaf data center design.

The topic of today's post is spine and leaf data center design.

I would like to highlight a bit about east-west and north-south traffic flow before going into more detail.

East-west traffic moves from server to server, while north-south traffic moves from client to server.

You guys must be aware of the access, distribution, and core layers. Please refer to the following picture

Here you can see the main building and building 1. Our main building could be used as a data center with our servers inside. In traditional data center environments, this traditional core, distribution, and access layer would work just fine where traffic flowed mostly north-south. North and south bound traffic flows are where traffic is flowing continuously up and down. There would be traffic going up and down the datacenter?and then to the clients in the other buildings. Through the distribution and access layer, it goes up to the core layer, and then down to the distribution and access layer. Using this model, you can see the flow of traffic from clients to servers to datacenters. Traditionally, campus designs work well when most of the traffic flows go north and south.

Today, we see a lot more eastward and westward traffic in modern datacenters. In the datacenter, east and west bound refer to the actual servers themselves. That's because data centers are getting better. Currently, there is a lot of virtualization. So many virtual servers and these servers might be clustered where an app is spread across multiple servers, and all of those servers need to communicate. An application might have a web-based front end on a server and a backend database on another server, for example. Again, the traffic would be going east-west between the different servers rather than north-south. The traditional campus design works well when most of your traffic is north and south bound, but not so well when it is east and west bound. Considering there is a lot of east-west traffic in modern data centers, there is another network design that is very popular now, and that is spine-leaf data center design. Refer the below diagram , you are probably looking at the diagram and thinking, oh, that's exactly like a collapsed core and distribution layer with the traditional model.?

Taking a look at the traditional model again, you can see that we pair up our distribution layer switches and our core layer switches. We have some load balancing and redundancy there as well, because we don't want to have a single point of failure. The spine-leaf design today looks the same, but it is actually designed in a way that allows for additional scalability and better performance for east-west traffic flows. As for the scalability, we can add additional switches in the east and west directions. As you can see from the below picture, I've got a larger data center, so I can just add more spine switches and leaf switches.

In the spine-leaf design, we have the spine switches here, which are at a higher level in the hierarchy. Our servers are connected to our leaf switches and we have a mesh between them. In this way, all of our leaf switches are connected to all of our spine switches, and it's really easy to scale this out by adding additional switches east and west. As a result, we have good scalability. Also, it gives us good performance, since if any of these servers in the data center need to talk to each other, they are only two hops apart. There will still be north- and south-bound traffic here, but it does give us those gains by adding east- and west-bound traffic.