Firefly - Scalable, Ultra Accurate Clock Synchronization for Datacenters
Open Compute Project via YouTube
Overview
Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about Firefly, a software-driven datacenter clock synchronization system that achieves sub-10ns device-to-device accuracy for cloud-based financial exchanges in this 19-minute conference talk by Google Software Engineer Yuliang Li. Discover how this innovative system addresses the critical challenge of maintaining Coordinated Universal Time (UTC) synchronization across datacenter infrastructure while overcoming limitations of existing clock sync techniques that struggle with clock drift, jitter, and path asymmetries. Explore Firefly's distributed consensus algorithm operating on a random overlay graph that enables rapid convergence to common time through gradual hardware clock adjustments. Understand the novel layered synchronization technique that decouples internal and external synchronization processes, allowing consistent sync-to-UTC across devices while maintaining stable device-to-device internal sync. Examine real-world performance results from a 248-machine Clos network deployment demonstrating sub-10ns device-to-device synchronization, ≤1µs device-to-UTC synchronization, and system resilience against time server failures and unstable clocks, making this solution both scalable and cost-effective for demanding datacenter environments.
Syllabus
Firefly Scalable, Ultra Accurate Clock Synchronization for Datacenters
Taught by
Open Compute Project