Assessing your IO, Memory & CPU loads is the first step - Knowing where the bottleneck is, is helpful for suggesting strategy. Often times joins are handy when you first create your DB as it pulls your data back in a tidy normalized result. As your data grows those joins can become troublesome because you are joining multiple large tables (meaning index lookups/loops through all the pertinent indexes), plus you can be bound to a single thread. Sometimes breaking that joined query into multiple queries that leverage separate threads as well as free up memory and the number of combined index's referenced at an instant between calls can be useful. I have not used Percona however, so it may have its own functionality that mitigates these concerns.