High-Speed Crypto Trading: JVM Techniques Behind Bitvavo’s µs Revolution

on YouTube ↗49m

" summary

Rebuild uses a deterministic replicated-state-machine on JVM with Raft, Aeron for reliable UDP + SBE + zero-copy transport, and strict zero-allocation code (object pools, flyweights, direct buffers). Measured internal processing hits 5 microseconds and end-to-end P99 stays under 1 ms (avg ~500 μs) while sustaining 100k orders per second. Deployment uses rolling upgrades with ingress snapshots and CI replay of the exact input log; kernel-bypass and CPU pinning eliminate OS jitter.

AI-written summary · about

" chapters (9)

1. 0:00Bitvavo and team intro
2. 7:24Startup to high-performance needs
3. 13:22Replicated state machine pattern
4. 20:02Networking with reliable UDP
5. 23:55Zero-copy SBE encoding
6. 27:27Safe distributed deployments
7. 34:08JVM low-latency best practices
8. 38:05Performance results
9. 40:14Q&A on networking trade-offs

" related talks

• 54mCrypto Trading: JVM Techniques Behind Bitvavo’s Microsecond RevolutionMarcos Maia, Oleg Lobanov · JCON EUROPE 2026
• 49m1BRC – Nerd Sniping the Java CommunityGunnar Morling, Roy van Rijn · Devoxx Belgium 2024
• 52mLow Latency Java in the Real World: LMAX Exchange & the Zing JVMGil Tene · GOTO Chicago 2015
• 51mJava at Speed: Getting the Most Out of Modern HardwareGil Tene · Oracle OpenWorld 2017
• 47mUltralow Latency with Java 11 and Terabytes of DataPer Minborg · Jfokus 2019
• 21mFaster Data With Faster JavaSimon Ritter · Big Data LDN 2022
• 45mProcessing 1m msgs/CPU/GB of memory with Quarkus and MicroProfile ...Loïc Mathieu, Victor Gallet · Devoxx France 2022
• 51mHigh Bandwidth Data & Low-latency JavaRyan Pratt, Justin Flude, Michael Vander Pluym · GOTO Chicago 2016