Cisco CCNP Data Center · DCCOR 350-601 · 1.0 Network · NX-OS

Data center foundations: spine-leaf, underlay and overlay, and NX-OS first steps

Why modern data centers are built as spine-leaf fabrics, what "underlay" and "overlay" mean, and how NX-OS differs from IOS — then configure your first Nexus uplink and OSPF adjacency.

62 min read10 chapters1 labs15 quiz8 scenarios15 interview Q&A

This first module is free: read the lesson and take the quiz. Labs need an account: sign up free to run up to 3 hands-on labs.

Log inStart free trial
This reading material cannot be printed. Read it on config.nwkings.com.
01

Module goals, prerequisites and the east-west recap

What you will learn in this module. This is the first real network module of the CCNP Data Center track. The beginner modules showed you what a data center is and refreshed the networking basics. Now you will learn how a modern data center network is actually designed and built. By the end of these ten chapters you will be able to:

  • Explain why the classic three-tier design struggles with modern traffic, and what vPC patched.
  • Draw a spine-leaf (Clos) fabric from memory, state its wiring rules and walk a packet through it.
  • Size a fabric: count ports, calculate oversubscription and decide how many spines you need.
  • Explain Layer-2 versus Layer-3 fabrics, and the underlay/overlay split behind VXLAN EVPN and ACI.
  • Recognise the Cisco Nexus families and choose sensible leaf and spine models.
  • Describe how NX-OS is built (features, processes, high availability, VRFs, licensing).
  • Configure your first Nexus: a routed /31 uplink, a loopback and an OSPF adjacency to a spine, then verify and save it.

Prerequisites

You do not need Nexus experience. You do need four things, all covered in the earlier modules:

  • IPv4 addressing, including what a /32 host route and a /31 point-to-point subnet are.
  • Switching basics: MAC learning, VLANs, trunks and why Layer-2 loops are dangerous.
  • Routing basics: what a routing table is, and the idea of an OSPF neighbor and area 0.
  • CLI comfort: moving between exec and configuration mode on a Cisco device.

If any of these feel shaky, revisit the dc-net-basics module first. Everything here builds on it.

Where this module sits in the track

dc-intro dc-net-basics dc-foundations you are here Underlay, vPC VXLAN EVPN ACI Storage, UCS, automation

This module gives you the vocabulary and the first configuration that every later network module assumes.

Every later module depends on this one. VXLAN EVPN is simply an overlay on the routed underlay you configure here. ACI builds the same spine-leaf shape and hides it behind a controller. Storage and compute modules plug into the leaves you learn to size. If the design ideas in this module are clear, the rest of the track feels logical instead of magical.

Quick recap: north-south and east-west

Draw the data center with the internet at the top. Traffic that enters or leaves the building travels up and down the drawing, so we call it north-south. Traffic that stays inside, from one server to another, moves sideways and is called east-west.

Twenty years ago most applications were a single big program on one server, and most traffic was north-south: a user asked, the server answered. Today an application is split into many small services (microservices), runs on virtual machines and containers that move between hosts, and reads from distributed storage and databases. One user click can cause dozens of internal calls. East-west is now the dominant direction in almost every data center.

Worked example: how big is east-west? An online shop handles 2,000 checkouts per second at peak. Each checkout sends one request in (north-south) and returns about 20 KB. Behind the scenes each checkout makes about 30 internal calls (cart, price, stock, payment, fraud, loyalty, logging), each moving about 50 KB.

North-south: 2,000 × 20 KB = 40 MB/s, about 0.3 Gbit/s.
East-west: 2,000 × 30 × 50 KB = 3,000 MB/s, about 24 Gbit/s.

East-west is roughly 75 times larger. Add backups, database replication and VM migrations, and the gap grows further. The network must be designed for the sideways traffic first.

Three design consequences follow directly from this recap, and you will meet each one again in the chapters ahead:

  • Equal distance. Any server should reach any other server in the same number of hops, so latency is predictable wherever the scheduler places a workload.
  • All links forwarding. Bandwidth you paid for must carry traffic; no link should sit idle waiting for a failure.
  • Small failure domains. A problem in one rack must not spread across the whole data center.

The lab you will build

SPINE1lo0 10.255.0.11 LEAF1lo0 10.255.0.1 Eth1/1 10.1.1.0/31 Eth1/1 10.1.1.1/31 OSPF UNDERLAY, area 0

The hands-on lab: one new leaf joins an existing spine with a routed /31 link and OSPF.

SPINE1 already runs an OSPF process called UNDERLAY. LEAF1 is a brand-new Nexus. You will enable the OSPF feature, turn Ethernet1/1 into a routed port with 10.1.1.1/31, create loopback0 with 10.255.0.1/32, join area 0.0.0.0 as a point-to-point link, verify the adjacency and save. Chapter 9 walks through every command. Chapters 2 to 8 explain why the fabric is built this way, so the commands make sense.

How to study this module

  • Read chapters 2 to 6 with a pen. Redraw every diagram yourself; the exam and interviews reward people who can sketch a fabric.
  • Do the sizing calculations in chapter 4 by hand before reading the answers.
  • Type the NX-OS commands in the lab rather than pasting them. Your fingers will remember no switchport and /31.
  • Finish with the scenarios and quiz. If you miss a question, go back to the chapter named in the explanation.

Common beginner mistake. Jumping straight to configuration and skipping the design chapters. You can copy an OSPF template in five minutes, but when a spine fails or a leaf runs out of uplinks you need to understand why the fabric is shaped the way it is.

Exam tip. The DCCOR 350-601 blueprint puts network design and NX-OS operation in its largest domain. Expect questions on spine-leaf rules, oversubscription maths, underlay versus overlay, VXLAN basics and NX-OS feature behaviour.

The nightly backup that slowed the whole row

A mid-size company ran a three-tier network. Every night at 01:00 the backup servers in row C pulled data from application servers in row A. Users on the night shift complained that the ERP system became painfully slow. The engineer checked the aggregation switch uplinks and found them at 98% utilisation, while the second uplink of each access switch sat idle because spanning tree had blocked it. The backup, a purely east-west flow, had to climb to aggregation and back down, competing with everything else. Short term, the team moved the backup window and enabled vPC so both uplinks could forward. Long term, the next refresh was designed as a spine-leaf fabric sized for east-west traffic.

Lesson: measure and design for east-west traffic first; it is usually the biggest load in the building.

"What do you understand by east-west traffic, and how would it change your network design?"

Strong answer: east-west is server-to-server traffic that stays inside the data center, such as microservice calls, database replication, storage and VM migration. It now far exceeds north-south traffic. That drives three design goals: every server at equal distance (predictable latency), every link forwarding (no STP blocking), and small failure domains. Those goals lead to a routed spine-leaf fabric with ECMP. Quote a quick number, such as one user request fanning out into 20 to 30 internal calls.

Key takeaways

  • This module teaches fabric design (three-tier, spine-leaf, sizing, underlay/overlay), Nexus hardware, NX-OS and your first routed uplink with OSPF.
  • You need IPv4 (/31, /32), switching, basic OSPF and CLI comfort before starting.
  • East-west (server to server) traffic is far larger than north-south in modern data centers.
  • Design goals that follow: equal distance, all links forwarding, small failure domains.
  • The lab joins LEAF1 to SPINE1 with a routed /31, loopback0 and OSPF UNDERLAY area 0.

Upgrade to keep reading

The free plan lets you read chapter 1 of this module. Upgrade to open the remaining chapters.

  1. The three-tier design and why it breaks
  2. Spine-leaf (Clos) fabrics from first principles
  3. Sizing a fabric: ports, oversubscription and scale
  4. Layer-2 fabric or Layer-3 fabric?
  5. Underlay and overlay
  6. Cisco Nexus hardware: families, chassis and ports
  7. How NX-OS is built: features, processes and high availability
  8. NX-OS for IOS engineers: your first routed uplink and OSPF
  9. Summary and exam checklist

Join the community