How would you approach a phased network upgrade when production has to stay online, the old and new networks can't be fully isolated, and the documentation and test environment aren't reliable?
How would you upgrade a legacy network that can't go down?
Answers
Where budget and space allow, build as much of the replacement alongside the existing network as you can. Test connectivity and application access before moving production traffic, then migrate a small site or service at a time rather than making one big cutover.
Keep the old path available for rollback until the new one has proved stable. Parallel hardware doesn't eliminate risk at shared connections, so those integration points still need careful testing and a clear backout procedure.
Reply to this answer
Start by finding out what is actually running. Capture current configurations, map the physical and logical connections you can verify, and identify the services that depend on each path. Treat old diagrams as clues, not proof.
Bring in the people who operate and use the affected services. They may know about dependencies that aren't obvious from a configuration file. Record the gaps too, because an unknown dependency is a reason to limit the size of a change.
Reply to this answer
No useful lab is a problem worth spending money on. Look for matching spare equipment and relevant modules where possible, then rehearse the upgrade and rollback repeatedly using configurations based on what you found in production.
Document the commands, expected results, failure points, and recovery steps. A partial lab won't reproduce every dependency or production load, so don't treat a successful rehearsal as a guarantee. It can still catch a lot before the real change.
Mavzelo+4r:
And test the backout itself, not just the forward upgrade. A rollback plan that only exists on paper may not work when the pressure is on.
Reply to this answer
Make each production change small enough to understand and reverse quickly. Pick a low impact candidate first, define what success looks like, and monitor the affected traffic and services before moving on. Schedule the work for a period of lower demand, with the right people available to respond.
Have technical, operational, and compliance stakeholders review the plan and the test results. If there is no credible way to test or roll back a step, that risk needs to be surfaced before anyone approves it.
Reply to this answer
Continuous uptime is a requirement, not a technique. If shared components can't be isolated and there's no reliable test or rollback, nobody can honestly promise a risk-free change. The project may need more equipment, time, or an explicitly accepted interruption.
Reply to this answer
Add your answer
Answers are reviewed before publication.
Zelmavo-2c:
That makes sense. The shared connections seem like the part where even a parallel build could affect production.