Cloud & Hosting
Moving to the cloud without taking the business offline
A staged approach to cloud migration for small and mid-sized systems: what to move first, how to test it, and how to keep a way back.
Codexan
6 min read
Cloud migrations fail in a predictable way: everything is moved in one weekend, something unexpected breaks on Monday, and there is no quick way back. The alternative is slower on paper and much faster in practice. Move in stages, and keep the old system working until the new one has earned your trust.
Start with an inventory
You cannot plan a move for systems you have not listed. For each application, record:
- what it does and who depends on it
- where its data lives and how large it is
- what it talks to: databases, file shares, other applications, external services
- how much downtime the business could tolerate
The surprises are always in the third line. The forgotten script that reads a shared folder at midnight is what breaks the migration.
Choose an approach for each system
Not everything deserves the same treatment.
- Retire. Some systems are no longer used. Switching them off is the cheapest migration there is.
- Replace. A self-hosted email or file server is usually better swapped for a hosted service.
- Move as it is. Copy the server to a cloud virtual machine. Fast, low risk, and no immediate savings.
- Rework. Change the application to use managed databases, storage and scaling. More effort, and the option with real long-term benefit.
A sensible plan mixes all four. Moving a system unchanged first, and improving it once it is stable in its new home, is often the least risky route.
Order the work
Begin with something that matters little: an internal tool, a staging environment, an archive. The first move teaches the team the networking, access and deployment details that every later move depends on.
Leave the system the business cannot do without until those lessons are learned.
Move the data carefully
Data is the hard part. For anything that changes during the working day:
- Copy the data to the new environment while the old system keeps running.
- Keep the copy up to date with continuous replication.
- Run the new application against it and compare results with the old one.
- At a quiet moment, stop writes to the old system, let replication catch up, and switch over.
Done this way, the visible interruption is minutes, not a weekend.
Keep the way back open
Until the new environment has run cleanly through a full business cycle, including a month end if that matters to you, keep the old system intact and know exactly how you would return to it. Write the steps down and decide in advance who can make that call.
Most migrations never use the rollback plan. The ones that need it and do not have one are the stories people tell.
After the move
A migration is not finished when traffic switches over.
- Check the bill after the first full month. Oversized machines and forgotten test resources are the usual culprits.
- Test a restore. A backup that has never been restored is a hope, not a backup.
- Review access. Remove the temporary permissions the migration needed.
- Switch off the old system on a set date, so nobody is left paying for both.
The goal is not to be in the cloud. It is to have systems that are easier to run, safer and cheaper than what you had before, and each of those can be measured.
- cloud
- migration
- aws
- azure
- planning