◆
No big-bang date
Workloads move individually. There is no single evening on which the company changes infrastructure.
Most migration failures are not technical. They are a consequence of a plan that required everything to move together, on a date, with no way back. We do not run that plan.
Used to move workloads off existing providers, data centres and legacy estates.
What this looks like in practice
Workloads move individually. There is no single evening on which the company changes infrastructure.
Parallel operation until you are satisfied. The decision to stop paying for the old arrangement is yours and comes last.
If a workload does not behave, it goes back. That is a Tuesday, not a crisis.
You decide based on what the workload actually did here, under real traffic.
The riskiest thing about a migration is usually the plan. A programme that must succeed entirely or fail entirely concentrates risk at exactly the wrong point.
The defining artefact of a traditional migration is a weekend with a runbook, a bridge call and a point of no return. Moving workloads individually removes that artefact entirely. There is no weekend, because there is no moment at which everything changes.
Running across both estates during transition, deliberately rather than as a failure state.
The part that genuinely is hard, handled with verification rather than optimism.
The old arrangement switched off when you choose, after it has been redundant for a while.
What the first workloads actually cost and did, as the basis for the rest.
We moved eleven workloads before anyone outside the team noticed a migration was happening. That was the goal and I did not expect to hit it.
The first workload is usually weeks. The whole estate takes as long as you choose, because nothing forces the pace once the first one is running and the old arrangement is still there.
No. Normal delivery continues, which is the main practical difference from a programme organised around a cutover date.
It goes back. Reversal is exercised as a routine step rather than documented as a contingency nobody has tried.
When you decide to, which is after the replacement has carried real traffic for as long as you want. Parallel running costs something and buys certainty; most teams judge the trade worth it.
Start on something that will still be right at a hundred times your size, and pay now what a startup should pay.
Get capacity when the experiment is ready, keep what produced a result, and put models in front of users without a separate platform.
Keep everything at full fidelity, query it without a project, and stop deciding today what you will want to ask next year.
Ship the application your users touch, on an address that stays put, with every version one click behind you.
Build the tools your own teams need, at a cost that makes small improvements worth doing.
Give your engineers a platform that removes decisions rather than adding them, without a platform team to run it.
Begin with one workload. Keep what you have running. Decide whether to continue once you have seen it carry something real.
No migration required to begin. Nothing to tear down if you stop.