I have Aurora cluster with a writer and several readers with autoscaling. CPU usage of my writer is constantly on 30-40% so I want to downgrade the instance to save money. Does anyone know how to estimate the CPU usage for my writer if I downgrade the instance from r5.4xlarge to r5.2xlarge, will the CPU be on double the numbers that were before of not? Any suggestions are welcome since the downgrade needs to be done on live environment.
Another thing to consider: when you go from 4xlarge to 2xlarge, the size of your buffer cache will also be halved. Which might mean that the "working set" of your application no longer fits into memory, causing extra CPU and I/O. One option for this case is to use x2g instance classes, which have the same CPU as the corresponding r5/r6 classes, but double the memory. (x2g classes are only available in some AWS Regions though.)
In this case, if you went r5.4xlarge -> x2g.2xlarge, you would reduce the CPU capacity by 50% but keep the memory the same. The cost would be somewhere in the middle, between r5.2xlarge and r5.4xlarge. Something to keep in mind if reducing to r5.2xlarge isn't practical because of the smaller RAM size.