Campus design
Campus networks are built in tiers: access, distribution and core
A three-tier campus has access, distribution and core layers; small sites merge the top two into a collapsed core.
Users, phones and access points plug into access switches, one or more per floor. Each has two uplinks. Distribution switches gather a building’s access switches, route between their VLANs and apply policy. The core does one thing: moves traffic between buildings as fast as it can. A fault in building A stays in building A.
Campus design · Going deeper
Topologies: star, mesh, point to point, hybrid
A full mesh needs n(n−1)/2 links: 10 for 5 sites, 190 for 20. Most networks are hybrid.
Data centre fabric
In a spine-leaf fabric, every server is the same distance from every other
Spine-leaf: every leaf links to every spine (32 links here), so east-west traffic always crosses leaf, spine, leaf over equal-cost paths.
In a data centre, servers plug into leaf switches, and every leaf connects to every spine. Here, 4 spines and 8 leaves. Server A on the first leaf talks to server B on the sixth: up to any spine, down to B’s leaf. Every pair of servers is the same three switches apart, and the traffic spreads across all spines at once.
Data centre fabric · Going deeper
VXLAN overlays
VXLAN wraps frames in UDP 4789 with a 24-bit VNI: 50 bytes of overhead, 16 million segments.
Scaling services
Load balancers spread the work; CDNs move content closer
A load balancer sends requests to healthy servers behind one virtual IP; a CDN serves cached content nearby: 800 ms becomes 32 ms.
Users connect to one address on the load balancer. It sends each request to a server that passes its health checks, and quietly skips the one that does not. Distance is the other problem. A user in Sydney loading a page from Virginia waits 200 milliseconds per round trip, 800 in all. A content delivery network keeps a copy in Sydney: 32 milliseconds.
SDN
Software-defined networks move the decisions into a controller
SDN separates the control plane from the data plane; a controller has a northbound API (REST) and southbound protocols.
A traditional switch does everything itself: you manage it one by one, it runs its own protocols to decide, and it forwards. With a controller, the deciding and the managing move to one place. Software talks to the controller through its northbound REST interface; the controller programs the switches through southbound protocols, and the switches just forward.
SDN · Going deeper
REST APIs and JSON
REST uses HTTP verbs (GET, POST, PUT, PATCH, DELETE) and returns JSON; tokens authenticate.
Automation
Automation turns one change into the same change everywhere
Ansible applies YAML playbooks over SSH without agents: 300 switches in 8 minutes, not 15 hours, and the same every time.
This playbook says one thing: every access switch uses this time server. By hand, 300 switches at 3 minutes each is 15 hours, and someone will make a typo. Ansible logs in to five at a time and is done in about 8 minutes. The playbook lives in Git, is reviewed like code, and can be run again to put back anything that drifted.
AI operations
An AI agent can run the operations loop, inside limits people set
Agentic AI observes, reasons and acts through APIs; good prompts give role, context, task, limits and format; guardrails keep changes approved.
An operations agent works in a loop. It reads telemetry and logs, forms a hypothesis, proposes a change with a way back, and for anything risky waits for a person to approve. Then it acts through the same APIs as your automation and checks the result. A useful prompt gives it a role, the context, a clear task, limits on what it may do, and the shape of the answer you want.
SD-WAN
SD-WAN sends each application over the best link right now
SD-WAN measures every path and steers each application by policy; SASE adds cloud-delivered security.
A branch has three ways out: MPLS, broadband and LTE. The edge router measures each one continuously. Voice takes MPLS, with twelve milliseconds and no loss; web traffic takes broadband; LTE waits as a backup. A central controller holds the policy for every branch, and if MPLS starts losing packets, voice moves within seconds.
Cloud networks
A VPC is a private network in the cloud, built from subnets and routes
A VPC holds public subnets (route to an internet gateway) and private subnets (out through a NAT gateway). An AWS /24 has 251 usable addresses.
A virtual private cloud is your own address space, here 10.0.0.0/16, split into subnets. The public subnet’s route table sends internet traffic to the internet gateway. The private one sends it to a NAT gateway. The load balancer and NAT gateway sit in the public subnet. App servers and the database sit in the private subnet: they can fetch updates, but nobody on the internet can reach them. The VPC connects home by VPN or a private circuit, and to other VPCs by peering or a transit gateway.
Cloud networks · Going deeper
Security groups, network ACLs and joining many VPCs
Security groups are stateful per instance; network ACLs stateless per subnet. 10 VPCs need 45 peerings or one transit gateway.
Lesson complete
You finished “Network design and the cloud”
Nicely done.