Database School A million transactions per second: building TigerBeetle with Joran Greef
16 snips
Oct 23, 2025 
Joran Greef, CEO and creator of TigerBeetle, dives into the groundbreaking world of high-throughput financial databases. He explores how TigerBeetle embeds double-entry accounting to enhance correctness and fault tolerance. Joran shares insights on the shift in transaction workloads, revealing how traditional databases face performance bottlenecks. He discusses innovative client-side batching for efficiency and explains the importance of resilience in database design. With a rich history of transactions, Joran illustrates why TigerBeetle stands out in modern commerce.
AI Snips
Chapters
Books
Transcript
Episode notes
Performance Puzzle In A Central Bank Switch
- Joran was hired to speed up an open-source central bank payment switch and found it capped at ~78 TPS.
- That investigation led him to realize general-purpose databases were the bottleneck for high-volume payments.
Amdahl And Contention Limit OLGP Throughput
- Using general-purpose SQL for high-contention debit/credit causes network round-trip locks and Amdahl-limited throughput.
- Real-world contention and serializability make OLGP databases hit practical limits (~100 TPS) under hot-key workloads.
Batch At Clients To Amortize I/O Costs
- Batch many small transfers at the client and send fixed-size numeric structs to amortize network and fsync costs.
- Let the database accept large numeric batches so it can fsync once and process thousands of transfers rapidly.
