An Evolution of a Betting Engine at William Hill
summary
Claims concrete scaling numbers 464 bets/sec, 5,000,000 price changes/day and 160TB/day and a working stream architecture built in Erlang with Apache Kafka and Docker microservices. Shows iterations: per-selection process model that hit messaging, supervisor and gproc limits, then a redesign using sharded ETS and short‑lived "bet flight" processes, plus Kafka persistence and Recon tracing to find a memory‑blowing accumulator bug; system runs at 500 cores and handles 500k price changes and 100k bets in 30 minutes.
AI-written summary · about
chapters (7)
- 0:00Start
- 1:39Bet Stream
- 10:08The Betting Engine
- 12:56Apache Kafka
- 24:25Bet Flow
- 41:29Identifying Suspicious Market Behavior
- 41:43Identification of Strange Bets
related talks
47mSports Betting Evolution
40mDistributed Computing challenges in the gaming & betting industry
52mThe Continued Evolution of High Frequency Trading Systems
37mBuilding a Scalable Real Time Bidding Exchange
35mMigrating from Kafka to RabbitMQ at SimpleBet: Why and How
27mScalable distributed systems from grounds up in Elixir
43mIntro to Stateful Services
38mMaintaining highly distributed services with Erlang: 20y experiences in fintech