As far as 'autonomic' DB tuning, it'd be nice to essentially have a model that provides a set query performance and redundancy for a lowest possible dollar amou

As far as 'autonomic' DB tuning, it'd be nice to essentially have a model that provides a set query performance and redundancy for a lowest possible dollar amount for a given workload. i.e I want my application's queries to have 99th percentile <=100ms latency - use whatever combination of memory, flash, disk, or compute is necessary to make that happen with four-nines reliability minimizing price. It'd then figure out what columns need to be stored on which medium, what in-memory indexes should exist, how many 'nodes' should be spot-instances vs reserved and so on..