Spine-Leaf Architecture
Last updated
Last updated
Data centers are dedicated spaces/buildings used to store computer systems such as servers and network devices.
Traditional data center designs used a three-tier architecture (Access-Distribution-Core) like we just covered.
This worked wel when most traffic in the data center was North-South.
With the precedence of virtual servers, applications are often deployed in a distributed manner (across multiple physical servers), which increases the amount of East-West traffic in the data center.
The traditional three-tier architecture led to bottlenecks in bandwidth as well as variability in the server-to-server latency depending on the path the traffic takes.
To solve this, Spine-Leaf architecture (also called Clos architecture) has become prominent in data centers.
There are some rules about Spine-Lead architecture:
Every Leaf switch is connected to every Spine switch.
Every Spine switch is connected to every Leaf switch.
Leaf switches do not connect to other Leaf switches.
Spine switches do not connect to other Spine switches.
End hosts (server, etc.) only connect to Leaf switches.
The path taken by traffic is randomly chosen to balance the traffic load among the Spine switches.
Each server is sparated by the same number of 'hops' (except those connected to the same Leaf), providing consistent latency for East-West traffic.
