Starting from Scale in the Design of the AT Protocol
a day ago
- The author, a local-first advocate, worked on SSB and Beaker Browser before joining Bluesky to build AT Protocol, a federated network for high-scale open computing.
- AT Protocol departs from local-first designs due to scale requirements, aiming to handle Twitter-level traffic (e.g., 6,000+ tweets per second) while preserving openness and user autonomy.
- High-scale systems decouple read workloads, replicate data from a canonical store to purpose-built secondary databases (e.g., graph, OLAP), and use logs (like Kafka) for durable, low-coordination sync.
- Bluesky applied the same data replication pattern to the public WAN by treating network challenges as similar to internal datacenter faults, adding trust through signed merkle trees and defensive APIs.
- The Personal Data Server (PDS) network replaces userdata clusters; writes are signed and replicated via pull-based sync streams, allowing apps to recover by replaying from PDS hosts.
- AT Protocol currently handles 500-600 events per second (peaks of 2,000), with over 3,000 PDS operators, and aims to keep the internet as an open information commons.