Instead of pointer-chasing through B-Tree node hierarchies, this index models the key distribution using streaming piecewise linear regression with strict epsilon error bounds.

Lookups compute positions directly on the CPU ALU before a local L1 cache search, reducing memory by 99% and cutting latency by 3x.

all 2 comments

sorted by: hot top controversial new old

That is… very neat. I want to explore this more, but I’m pretty sure this could be super useful with some of the parquet stuff my team does

  • source
  • [–] 1 point 3 weeks ago

    The paper is a bit older, and I haven't seen this anywhere else. What's the catch?

  • source