DoorDash reports that peak demand approached four times the processing capacity of its legacy dispatch engine, whose algorithms had roughly quadratic time complexity. It says the redesign split the system into a stateful coordinator that manages global logistics state and business rules, and stateless compute workers that run routing heuristics and model inference. Mechanism: The coordinator retains global state and business-rule responsibility while distributing localized payloads to stateless workers for routing heuristics and inference.
The company reports tenfold higher throughput and 99.9% of runs under 20 seconds, alongside improved on-time delivery performance. Why now: The account describes peak demand making the previous execution model fall behind changing real-world conditions, so dispatch decisions could be stale before they were emitted. First implication: The reported architecture makes coordination and compute independently scalable, but its value depends on preserving enough local context for useful optimization. This is a company-authored case study, not independent verification; Evidence boundary: The supplied source reports outcomes but does not establish an external benchmark or disclose enough methodology to compare them independently.
First watch condition: Look for independently described workload definitions, measurement windows, and evidence that the throughput and latency gains persist across peak conditions.