This is interesting - compute/compute separation in Rockset: https://rockset.com/blog/introducing-compute-compute-separation/ This kind of architecture - where the DB is split into components that can then scale independently is very "cloud native" IMO. Makes no sense if you just need some analytics (use DuckDB in that case), but is very efficient to run at scale if you are a SaaS provider. WDYT?

agreed. https://yourdatafitsinram.net/ applies heavily. The only time I'm really interested in any separation/distributed system, is it makes technical sense(normally this means multi-tenant)

exactly