I've found that pgtune doesn't quite provide the benefit that I would like. It still uses large work mem and maintenance work mem numbers, even though, up until now Postgres has an issue with large numbers of tuples, so seems that smaller settings are better, in the 64MB type range. (based on feedback from this list in the past and testing of larger numbers on dedicated systems).
Also I've found no benefit to larger Effective cache numbers in boxen dedicated to postgres.
So looking for a good start as I start bringing up systems in Amazon, AWS for performance.
pgtune is a great idea, but it's numbers seem to be based on what should be, vs what is..
I'm currently on CentOS6 and 9.4.5
Hardware specs of the AWS systems are 8 cpu/60 GB, I may bump that to a 16/122gb, but trying to control costs and I know going from 8 to 32 yielded almost 0, my biggest gain was memory but even then I don't think I've got settings correct.
