Can local-first really scale at the edge?
summary
Local‑first faces linear metadata blowup when many replicas interact; this work surveys concrete workarounds and prototypes. Causal histories + content or update hash chains trade space for authenticity; probabilistic Bloom‑filter forgetting limits metadata growth but admits false negatives. For counters and sets, an atomic transfer protocol moves local replica IDs into a smaller server namespace to compress causal contexts; a Rust implementation for add‑wins sets shows much smaller on‑disk metadata versus standard CRDTs, though tree schemes need static topologies and some protocols remain probabilistic or brittle.
AI-written summary · about
chapters (9)
- 0:00Local-first scaling challenges
- 3:35Metadata costs in local-first
- 10:19CRDT counters determinism
- 15:42Scalable CRDTs for sets
- 19:19Implementation trade-offs
- 22:10Referee-based coordination
- 24:40Algorithm choices for topologies
- 28:54Byzantine behavior handling
- 31:00Q&A
related talks
24mGoing Local-First: A Primer
24mStronger Semantics for Low-Latency Geo-Replicated Storage
36mConflict-Free Replicated Datatypes (CRDT)
33mProgramming Support for Local-First Software
35mLocal-First Distributed Configuration
18mRethinking Apps with Local-First Architecture
32mAn Uncoordinated Approach To Scaling (Almost) Infinitely
50mImmutable Values in the Big Wide World