From the perspective of the development trend of data centers, the trend of "cloud" and digitalization of enterprises is irreversible. The demand for data centers in information and communication technology (ICT) industries such as cloud computing, big data, and artificial intelligence is increasing, and the demand for data center bandwidth is increasing. With rapid growth of 25% to 35% per year, bandwidth capacity has also doubled. The following will talk about the advantages and solutions of the Spine-Leaf network structure.

As a key part of new infrastructure, data centers must consider deploying 25G/100G and above high-bandwidth networks. With the maturity of standards and the improvement of technology, the commercialization of 25G/100G high-speed optical modules provides feasibility for the deployment of leaf-spine networks in data centers. Compared with the traditional three-tier hardware architecture, the leaf-spine architecture is more in line with the development trend of the data center. Its advantages are reflected in the following points:

1. Balanced load sharing, the uplink of each leaf switch works in a load-balanced manner, making full use of the bandwidth.

2. Estimated delay value, the number of connection paths between leaf switches can be determined, all need only pass through one spine switch, and the east-west network delay can be predicted.

3. Good scalability, most switches supporting 25G/100G can be backward compatible with 10G/40G, providing higher flexibility for the ports of the switch, and more flexible network expansion.

4. Reduce the performance requirements of the switch. North-south traffic can go out from leaf nodes or spine nodes. East-west traffic is distributed over multiple paths. This eliminates the need for expensive, high-performance, high-bandwidth switches.

5. High security and availability. In the leaf-spine architecture, when a device fails, there is no need to re-converge, and the traffic continues to pass on other normal paths. The network connectivity is not affected, and the bandwidth is only reduced by the bandwidth of one path. Performance The impact is minimal.

When designing under the leaf-spine architecture, more than 90% of the fiber links in small and medium-sized data centers are less than 100 m in length, and the link deployment scheme is mainly based on multimode fiber, and the optical module deployment is mainly 40/100G.

To form a 25G/100G Spine-Leaf network, you need to use 25G servers, 25G switches, 100G switches, as well as 25G SR optical modules and 100G SR4 optical modules from ETU-Link, OM3 optical fiber patch cord, distribution frames, etc.

To connect the 25G server and the 25G switch (downlink port), you can use 25G SR optical module with OM3 multimode fiber jumper, and to connect the 25G switch (uplink port) and 100G switch, you can use 100G SR4 optical module with OM3 multimode fiber patch cord. (The connection diagram as below)

The new leaf-spine network adopts a flat and large two-tier architecture, which expands the access and aggregation layers, greatly improving network efficiency, and does not need to purchase high-performance core switches, and is composed of relatively small-scale low-end switches. , mainly based on 40G/100G bandwidth deployment (large data centers can consider forward-looking deployment of 200/400G bandwidth), supports full-speed east-west traffic, and can reduce enterprise deployment and maintenance costs, providing a practical and effective development for data centers strategy.