Hi everyone. I'm in the middle of planning a lift-and-shift migration for a few on-premise legacy servers to AWS, and I'm hitting a wall with pricing estimates.
Every time I look up benchmark data or pricing guides from even six months ago, the numbers feel completely off. Instance types get deprecated, new ones pop up, and reserved instance discounts seem to change constantly. It makes it impossible to get a solid budget approved when I can't trust the baseline numbers I'm finding.
How are you all dealing with this? Are there any reliable, *current* sources you use for cloud migration cost modeling? I'm especially nervous about database migration (SQL Server to RDS) and long-term storage costs. I just need a realistic picture so I can build a step-by-step plan and timeline that won't get shredded by finance later 😅
Any tips on where to find fresh data, or how you've calibrated your own quotes recently, would be a huge help.
One step at a time
You've absolutely nailed the main headache with cloud cost planning - it's a moving target. The "six month old data is useless" feeling is totally valid, especially around instance families.
What my team does is lean heavily on the cloud providers' own pricing calculators, but with a twist. We run the numbers, then immediately factor in a 10-15% "volatility buffer" for the first year. It's not perfect, but it gets finance comfortable. For SQL Server to RDS specifically, don't just look at instance costs - the real budget killers can be the licensing (BYOL vs. license-included) and backup retention. Those are often the numbers that slip through.
Have you considered running a small-scale proof-of-concept migration first? The actual bill from a month-long test of your most critical server is the only data point I fully trust. It gives you something concrete to scale from.
I completely agree that the PoC approach is the only way to get trustworthy numbers. That volatility buffer user649 mentions is crucial. We've adopted a similar practice, but we treat the first month's actual bill from a test as the new baseline, then apply the buffer on top of that for the full-scale projection. It creates a two-stage model that feels much more defensible.
One nuance with the SQL Server licensing point: the decision between BYOL and license-included can also lock you into a specific cost trajectory for years, which is a different kind of risk beyond just the initial estimate. Once you choose a path at migration, it's often prohibitively complex and expensive to switch later, so that test should explicitly compare both licensing models under real workload.
Method over hype