Free course · 10 lessons · Advanced
Cloud Networking in Production
The network problems that cause real outages in the cloud: overlapping address ranges, exhausted subnets, NAT port exhaustion, MTU black holes, stateless ACLs, slow DNS cut-overs, 502s from keep-alive timeouts and health checks that take everything down. AWS first, with Azure and Google Cloud equivalents, runnable simulations in every lesson and two capstone case studies.
1. Address Planning
Most network redesigns start with an address plan that was too small or overlapped something. Plan ranges so networks can grow and connect.
- The Acquisition That Couldn't Connect — Two networks with the same private range can't be peered or routed. Plan ranges centrally, and know the workarounds when you inherit an overlap.
- Pods Stopped Scheduling — Kubernetes and serverless workloads can consume a subnet's addresses fast. Size subnets for them, and use prefix delegation or secondary ranges.
2. Paths That Break
Connections that fail only sometimes, only for big responses or only for some clients: NAT ports, MTU and stateless rules.
- Timeouts to One Partner API — A NAT gateway can only open so many connections to one destination. Reuse connections, add NAT addresses, and watch the port-allocation metric.
- Large Responses Hang Over the VPN — When ICMP is blocked, Path MTU Discovery fails and big packets vanish. Allow the right ICMP, clamp MSS on tunnels, and test with real packet sizes.
- The Network ACL That Broke Some Clients — Stateless network ACLs must allow return traffic to ephemeral ports, and different clients use different ranges. Prefer security groups.
3. Names, Load Balancers and Timeouts
DNS caches, keep-alive timeouts and health checks decide whether changes and failures are invisible or become outages.
- The Migration That Took Two Days — DNS changes are only as fast as the longest cache. Lower TTLs in advance, shift traffic with weighted records, and know which clients cache on their own.
- Random 502s Behind the Load Balancer — When the app closes idle keep-alive connections before the load balancer does, some requests land on dead connections. Keep the app's timeout longer.
- The Health Check That Took Everything Down — Health checks that test shared dependencies mark every instance unhealthy at once. Keep load balancer checks shallow and about the instance.
4. Capstone Case Studies
Design a multi-account network for a growing company, and investigate intermittent checkout timeouts.
- Capstone: A Network for a Growing Company — Twenty AWS accounts, two offices, a data centre and an Azure deployment on the way. Design the network so it can grow and stay secure.
- Capstone: Postmortem — Intermittent Checkout Timeouts — Reconstruct an incident from metrics and configuration, find every contributing cause, and write the remediation plan.