I'm not sure if its completely favourable for cloud sql to perform worse than aurora when the thread count goes above 16. Seems like two different use cases are being targetted:
Aurora for consistent performance as concurrent users/queries increase (but sacrificing speed at the small scale). Cloud sql for high performance on less concurrent workloads.
They both perform significantly faster than RDS MySQL. That's probably the lesson here.
SQL connection pools are configurable in the application. When you care about performances, you have to do benchmarking to find the optimal number of concurrent threads for the specific load.
It's noteworthy to see that google has a consistent total of queries per seconds (the average per thread drops almost linearly when adding threads), whereas RDS Aurora is exhibiting variable throughput and flat areas.
