Series: Cisco Data Center Spine and Leaf Architecture – Design Overview Whitepaper

Series: Cisco Data Center Spine and Leaf Architecture – Design Overview Whitepaper

Data centers are the foundation of modern software technology and play a crucial role in enterprise scalability. Traditional data centers use a three‑tier architecture, with servers grouped into pods as shown in Figure 1.

Figure 1. Traditional three‑tier data center design

This architecture consists of core routers, aggregation (or distribution) routers, and access switches. Between aggregation routers and access switches, the Spanning Tree Protocol (STP) builds a loop‑free Layer 2 topology, offering plug‑and‑play simplicity and VLAN extension within each pod, but it cannot use parallel forwarding paths and always blocks redundant VLAN links.

In 2010, Cisco introduced Virtual Port Channel (vPC) technology to overcome STP’s limitations. vPC eliminates STP‑blocked ports and provides active‑active uplinks from access switches to aggregation routers, fully utilizing available bandwidth, while STP remains as a fail‑safe mechanism.

vPC works well in relatively small data center environments where most traffic is north‑south between clients and servers.

Figure 2. Data center design based on vPC

Since 2003, with the introduction of virtualization, compute, network, and storage resources isolated in Layer 2 pods can be aggregated, creating a larger Layer 2 domain from the access layer to the core, as illustrated in Figure 3.

Figure 3. Data center design with an expanded three‑tier domain

Extending Layer 2 segmentation across all pods allows administrators to create a centralized, flexible resource pool that can be re‑allocated as needed. Servers are virtualized into a set of virtual machines that can move freely between physical servers without changing operating parameters.

With virtualized servers, applications are increasingly deployed in a distributed manner, leading to higher east‑west traffic that requires efficient handling with low and predictable latency. However, vPC provides only two active parallel uplinks, making bandwidth a bottleneck in three‑tier architectures, and server‑to‑server latency depends on the chosen communication path.

To overcome these constraints, a new data center design based on a Clos network spine‑and‑leaf architecture has been developed. This architecture has been proven to deliver high bandwidth, low latency, and non‑blocking server‑to‑server connections.

Original source: Cisco Whitepaper

Article: http://jiagoushi.pro/node/1031

Discussion: Join the Knowledge Planet "Chief Architect Circle" or add the author’s WeChat ID "intelligenttimes".

WeChat Public Account

Follow the public account

Chief Architect Think Tank

WeChat Small Account

Groups to join: Architecture, Cloud Computing, Big Data, Data Science, IoT, AI, Security, Full‑Stack Development, DevOps, Digitalization, Product Transformation.

Knowledge Planet

Ask experts, get close interaction, or receive private sharing.

Click to join Knowledge Planet "Chief Architect Circle"

WeChat Circle

Exchange with like‑minded peers.

Click to join WeChat Circle "Chief Architect Circle"

Himalaya

Listen to the latest tech news and architecture insights while traveling.

Click to listen to "Smart Moments, Architecture Talk"

Knowledge Planet

Meet more friends, discuss careers and technology.

Click to join Knowledge Planet "Knowledge and Technology"