Do the following:
- Reproduce the performance problem on non-production hardware, but production-grade hardware in your lab
- Upgrade to Percona server, or carry out other changes, and measure performance to find out whether it fixes it.
- Assuming your results are good, carry out MASSIVE FUNCTIONAL REGRESSION TESTING on your application, to check that there are no regressions introduced by the move to Percona server.
- Carry out Soak and stress testing on your application - on your non-production system with production-sized data (or bigger).
- Test the roll-back plan, to verify that you can actually successfully go back to MySQL
- If you are TOTALLY happy with the results of ALL OF THE ABOVE, then you can plan the release to production.
But honestly, if you don't want to do those things, consider finding another way to fix it. Maybe there is a way you can fix your application to store fewer data, modify data less often, etc.
Perhaps you can add more powerful hardware?
10 million records is absolutely tiny and even the smallest of servers (say, 16G ram) should be able to keep the entire DB in memory. So unless you're doing massive amounts of updates,
Can you not find a less risky way of doing it?