I scanned this, and there's a lot of talk here about having two separate layers of key-value storage (not nearly the first system to have that design), but I'm missing what is the key innovation here. I need to read it in more detail, but if anyone else figures out what that is, I'd love to hear about it.
Same here, can't see what's the discovery in this paper.
Or normalize your data and use clickhouse.
Clickhouse is a nightmare to scale and operate and doesn't support TS joins.