Distributed Sort Design — TeraSort at Scale

Staff+ interviews sometimes ask the pure systems question: 'Sort 1 TB of data distributed across 1000 machines, each with 1.5 GB RAM, as fast as possible.' This is not a coding question — it's a distributed systems design question where sorting is the core primitive. The right answer shows you understand range partitioning, parallel sort, and fault tolerance. The Range Partitioning Algorithm Tournament Tree (Loser Tree) — Optimal K-Way Merge