Serviceability under high demand
summary
Defines a concrete metric stress ratio = undelivered orders / delivery agents and treats problem when s > 1; models the undelivered orders as a queue, uses short‑horizon (≈10 minute) predictive models and a quasi‑static assumption for delivery rate, and enacts real‑time throttling by shrinking restaurant visibility (shrink radius) or reserving slots per customer segment. Assignment logic prioritizes percentile targets (e.g. 90th) and batching rules; estimating the counterfactual unconstrained flow proves hard for A/B tests. Claims these measures measurably raise throughput and reduce cycle time.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:11Swiggy's Dream Scenario
- 2:12REAL WORLD CONSTRAINTS
- 3:55How do we address them?
- 9:08QUANTIFY DELIVERY SYSTEM LOAD
- 11:36QUEUEING MODEL
- 13:48ORDER RATE, STRESS AND LAG
- 16:03DEMAND PREDICTION
- 19:23DEMAND SHAPING
- 25:14Order loss
- 26:45Restaurant parameters
- 28:18ASSIGNMENT ALGORITHM
related talks
38mA Practical Look at Performance Theory
27mService Degradation at Scale: Creating Instant Capacity
56mDefending against CDD: Chaos-Driven Delivery
30mThe Queue Continuum: Applied Queuing Theory
35mCapacity Planning in Four Parts: Telling the Future without a Crystal Ball
41mSLOs that Delight: Master Multi-Service SLO Calculations
05mCombat Chaos-Driven Delivery By Thinking Like An OS
31mWhat does "high priority" mean? The secret to happy queues