IT Just Killed B-Trees — I Spent 2 Weeks Proving They’re Right (And Wrong)

A 50 year old data structure just got outpaced by a radical design, but the truth I found in the benchmarks surprised me more than the hype.

The cursor blinked on a terminal window at 11 PM. Two identical queries sat in the editor, one against a B-tree index, the other against MIT’s new Adaptive Radix Tree implementation. The fan hummed. I hit Enter twice, watched the milliseconds stack up, and felt something shift. Forty seven years of database orthodoxy, suddenly negotiable.

You tell yourself the fundamentals are settled. B-trees won decades ago. Every database you’ve touched, Postgres, MySQL, MongoDB, leans on them. Then a paper lands, benchmarks attached, claiming 3 to 5 times faster lookups with half the memory overhead. And you’re left wondering if you’ve been cargo culting tree structures your entire career.

You’re not alone. A 2025 survey from Carnegie Mellon found that 68% of backend engineers still design indexes around B-tree assumptions without questioning whether the hardware underneath has changed enough to matter. Spoiler: it has.

Here’s what I found after two weeks of testing, breaking, and rebuilding those benchmarks, and why the…