Skybridge: Bounded Staleness for Distributed Caches
summary
Skybridge is an out-of-band replication service that indexes write metadata and answers cache staleness queries so caches avoid needless refills. It uses leases, Bloom filters, proactive streaming, and a relaxed replication with gap detection semantic that accepts bounded, detectable data loss and out‑of‑order delivery to favor low latency. On Meta production traffic it proves about 59% of reads fresh without refills, targets a 2-second bound, boosts consistency from ~four nines to five–seven nines, and has a 0.54% server footprint.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:43Bounded staleness importance
- 4:31Bounded staleness challenges
- 6:36Tracking write metadata
- 7:42Replication with gap detection
- 11:15Implementation trade-offs
- 12:39Effectiveness evaluation
- 15:10Conclusion
related talks
26mCache Strategies with Best Practices
30mBeacon: Intelligent Latency-Aware and Load Shedding Service Routing
15mBuilding Self-Healing Cache Infrastructure
18mScaling APIs to Billions: Extreme-Scale Reliability
25mDistCache: Provable Load Balancing for Large-Scale Storage Systems with Distributed...
47mCaching beyond simple put and gets
37mCaching Deep Dive: The Ultimate Way To Speed up Your GraphQL API
31mSolid Cache: A disk backed Rails cache