Comparing Two-Tier and Three-Tier Data Center Networks
In This Article
Modern applications require a modern data center infrastructure, and two-tier spine-leaf architectures offer many advantages over traditional three-tier designs.
More than 70 percent of all traffic today moves from server to server, or what we consider east to west traffic. Modern applications require significantly more data to travel within a data center at faster speeds and are less forgiving about latency. However, traditional data center networks were initially designed for resiliency and were concerned with speed into and out of the data center, not within it.
To solve this, we have been recommending that customers move to a two-tier or spine and leaf architecture in their data centers for the past several years. We can gain a better understanding of modern data center network considerations by comparing various tiered network architectures and how they solve modern business challenges.
Tiered architecture is a well-established approach to logically organizing switches within the data center. It determines how switches are cabled, the amount of redundancy, throughput, scalability, and opportunity for undesirable looping. Switches can be organized in multiple pods consisting of multiple tiers of switches. The traditional approach to tiered architecture is three-tier data center networks.
What are three-tier data center networks?
Traditional data center networks utilized a three-tier design that consists of a core, distribution and access layer of switches.
- Core switches are usually large modular chassis with very high throughput and advanced routing capabilities.
- Distribution layer switches are mid-tier speed switches with an emphasis on uplink speeds. Services, such as load balancing or firewalls, can often be found at this layer.
- Access switches are the traditional top-of-rack (TOR) switch that regularly consists of 24 to 48 ports of 1 or 10Gbps speeds with similarly sized uplinks.
Three-tier data center networks were the generally recommended data center designs in the past. They worked very well when most of the traffic moved north to south, from outside the data center in, or vice versa. A packet flows to the core, is routed to the correct distribution switch, and then forwarded to the access switch where the server was connected. Most legacy packets had traditionally moved through three physical hops, which limits the amount of latency added per-packet flow.
The main issue with this design for the modern data center is that intra-DC traffic is the new norm. Due to server-to-server traffic, three hops now quickly become four, five or more, adding significant latency per flow as well as increasing the risk of bottlenecks, buffer overruns and dropped packets.
Three-tier data center networks introduced loops, which you can see in the graphic above, required the correct spanning-tree protocol configuration. Spanning-tree issues are notorious for causing network outages as a spanning-tree failure causes continuous looping.
What are two-tier data center networks?
In the modern data center, we recommend two-tier spine and leaf architectures, also known as Folded-CLOS. This approach is better suited to meeting the needs of modern applications, such as high-throughput and low-latency.
- Spine switches are very high-throughput, low-latency and port-dense switches that have direct high-speed (40-400Gbps) connections to each leaf switch.
- A leaf switch is typically used as a TOR switch. Leaf switches are very similar to traditional TOR switches in that they are often 24 or 48 port 1, 10 or 40Gbps access layer connections. However, they have the increased capability of either 40, 100 or 400Gbps uplinks to each spine switch.
The more modern two-tier spine and leaf architecture approach offers a wide range of benefits, including:
- Resiliency: Each leaf switch connects to every spine switch. Therefore, a spanning tree is not needed and, due to TRILL, SPB or SDN protocols, every uplink can be used concurrently.
- Latency: There is a maximum of two hops for any east to west packet flow, so ultra-low-latency comes as standard.
- Performance: True active-active uplinks enable traffic to flow over the least congested high-speed links available.
- Scalability: You can increase leaf switch quantity to the desired port capacity and add spine switches as needed for uplinks.
- Adaptability: Multiple spine and leaf networks across a multicloud ecosystem can be connected and managed from a single pane of glass. Also, this topology has benefits in other areas of the enterprise network, such as industrial cell architecture or corporate LAN.
What should I consider when using two-tier spine and leaf architectures?
With a two-tier network architecture, the data center will need to be re-cabled to connect each leaf to each spine. This new design requires a considerable amount of cabling and optics for connectivity.
Two-tier spine and leaf architectures may still require core switches for layer three routing. Planning the physical and logical network is critical before purchasing the hardware for a new data center.
Modern applications require a modern data center infrastructure design, so two-tier spine and leaf architecture offers many advantages over traditional three-tier architectures. A two-tier network architecture eliminates single points of failure, traffic bottlenecks and scalability issues, as well as improving overall throughput and ease of management.
The two-tier architecture addresses the need to modernize the physical network. To do so, WWT recommends adding a software-defined network (SDN) platform like Cisco ACI or VMware NSX. You can learn more about transitioning data center infrastructure in one of our latest articles on intent-based networking or by scheduling an on-demand Cisco ACI lab, which explains how ACI helps you to set up your network in an automated manner.
Ready to get started? Your WWT account manager can bring the right experts together to help design your modern data center. Discover how WWT can help your organization build a strategy for next-generation network architecture.