What Is a Data Center Switch? ToR, EoR and Spine-Leaf Architecture

Sunucun 4 min read

Short answer: A data center switch is a high-port-density, low-latency network switch that connects the servers, storage systems and other network devices in a data center to each other and to the internet. The most common type is the ToR (Top of Rack) switch: it sits at the top of each rack and the servers in that rack connect directly to it. In today’s data centers, ToR switches are in turn usually interconnected with a spine-leaf architecture.

Illustration: What Is a Data Center S…

What Sets a Data Center Switch Apart from an Office Switch?

  • Port speed: server connections are usually 10, 25 or 100 Gbit/s, and links between switches (uplinks) 100 or 400 Gbit/s. Office switches typically run at 1 Gbit/s.
  • Low latency and high capacity: designed so that all ports can run at full speed at the same time.
  • Redundancy: dual power supplies, replaceable fans; with servers connected to two switches at once (methods such as MLAG), a single switch failure doesn’t drop the connection.
  • Rack-friendly cooling: front-to-back (or back-to-front) airflow to match hot/cold aisle layouts.
  • Transceiver modules: ports are module slots such as SFP28, QSFP28 or QSFP-DD, used with fibre or copper modules.

ToR, EoR and MoR: Where Does the Switch Go?

PlacementHow it worksAdvantageDisadvantage
ToR (Top of Rack)One (usually two, for redundancy) switches at the top of each rack; servers in the rack connect with short cablesShort cables, tidy cabling, grow rack by rackA switch per rack, more devices to manage
EoR (End of Row)Large switches in a rack at the end of the row; every server in the row is cabled thereFewer switches, central managementLong, crowded cable runs
MoR (Middle of Row)Same idea as EoR, with the switch rack in the middle of the rowShorter longest cable than EoRStill heavy horizontal cabling

In modern data centers with growing server counts and speeds, ToR is the most common approach. The short answer to “what is a ToR switch”: the first switch at the top of the rack that the servers in that rack connect to.

Classic Three-Tier vs Spine-Leaf Architecture

Enterprise networks long used a three-tier access – distribution – core design. It was built for networks where most traffic flows from users to the internet (north-south). In data centers, most traffic flows between servers (east-west): application servers talk to databases, virtual machines to storage. That’s why spine-leaf became common in data centers:

  • Leaf switches: usually the ToR switches in the racks; servers connect to them.
  • Spine switches: every leaf connects to every spine; spines don’t connect to each other or to servers.
  • Result: the path between two servers in different racks is always leaf → spine → leaf, so latency is predictable. Traffic is spread across multiple spines over equal-cost paths (ECMP), and the network keeps working if a spine fails.
  • Growth: add a leaf for a new rack, add a spine for more capacity.

What to Look for When Choosing a Data Center Switch

  1. Port count and speed: number of servers in the rack × link speed, plus future growth.
  2. Uplink capacity and oversubscription: the ratio of total server port speed to uplinks (e.g. 3:1). The lower the ratio, the lower the risk of congestion.
  3. Redundancy: dual power supplies, dual ToR and the ability to connect servers to two switches (MLAG, EVPN multihoming and similar).
  4. Software features: support for protocols such as VLAN, VXLAN/EVPN and BGP, and automation options (API, Ansible).
  5. Airflow and power: airflow direction that fits the rack layout, and power consumption.
  6. Licensing and support: some features need extra licences; check the vendor’s support period.

Frequently Asked Questions

Is a ToR switch the same as a leaf switch?

Often, yes. ToR describes the physical location (top of the rack), while leaf describes the role in a spine-leaf architecture. The switch at the top of a rack is usually also a leaf switch.

Why put two ToR switches in one rack?

For redundancy. Servers connect to both switches with one cable each; if one switch fails or is being upgraded, servers keep running through the other.

Do I need a data center switch for my own server?

No. When you rent a VDS or dedicated server hosted in a data center, the provider builds and manages the network. If you place your own hardware in a data center, network connectivity comes with the server colocation service.

Conclusion

Data center switches are the nervous system of the data center. ToR switches connect servers at rack level, and spine-leaf architecture ties those switches together with predictable latency and easy growth. If you’re new to networking concepts, start with our what is a network article.

Categories

Latest guides