Time Synchronization using ML Techniques in Electronic Trading
summary
Claims nanosecond-level synchronization using Stanford’s Hygens algorithm that runs in NICs, sends probes to ~10–20 peers and synchronizes every 2 seconds. It removes noisy measurements with coded probes, fits bounds with support vector machines, and resolves loop inconsistencies with a network-effect minimum‑norm solve. Tests report ~12–15 ns average and 30–45 ns 99th percentile at 40% load, ~16–23 ns average and 45–48 ns 99th at 90% load, using ~0.5% bandwidth and <4% CPU on a 40 Gbps fabric.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 3:10Importance of time sync
- 7:51Hygen's algorithm overview
- 16:47Comparison with protocols
- 19:16Performance results
- 20:19Conclusion and impact
related talks
26mExploiting a Natural Network Effect for Scalable, Fine-grained Clock Synchronization
46mSelf-Programming Networks: Network Clock Synchronization, Event Playback and Real-time Action
17mHow to diagnose nanosecond network latencies in rich end-host stacks
26mSoNIC: Precise Realtime Software Access and Control of Wired Networks
14mSyncWise: Error-Aware Time Synchronization for Reconfigurable Data Center Networks
19mAI/ML Networking Challenges: The Fast and the Finicky!
42mUsing the Network as a Reliable Platform for Time-Sensitive Systems
34mDeterministic Networking for Real-Time Systems (Using TSN)