Crypto Trading: JVM Techniques Behind Bitvavo’s Microsecond Revolution
summary
Bitvavo rebuilds its core on the JVM and Kotlin using a replicated state machine (Raft) with Aeron for reliable UDP and SBE for zero-copy messaging. The system removes Redis/MySQL from the hot path, uses preallocated buffers and object pools to avoid GC, and pins threads with optional kernel-bypass networking. Internal processing averages about 5µs, end-to-end P90 is ~500µs and throughput hits ~100k orders/s, proving the design works for deterministic, ultra-low-latency trading.
AI-written summary · about
chapters (10)
- 0:00Introduction and background
- 13:10Replicated state machine concept
- 21:14In-memory state and data structures
- 23:01Binary encoding with SBE
- 24:51Reliable UDP with Aeron
- 29:45Current system architecture
- 31:49Guardrails for production stability
- 37:12JVM memory and concurrency optimizations
- 42:32Performance benchmarks and results
- 44:39Q&A session
related talks
49mHigh-Speed Crypto Trading: JVM Techniques Behind Bitvavo’s µs Revolution
52mAchieving Sub‑Millisecond Response Times in a Java Trading System on Linux
01h 01mJava compilers: Why Java is one of the first options for Low latency apps
52mLow Latency Java in the Real World: LMAX Exchange & the Zing JVM
47mUltralow Latency with Java 11 and Terabytes of Data
56mModern Java microservices in the Cloud!
49m1BRC – Nerd Sniping the Java Community
35mJava at Speed: Building a Better JVM