Systems and Technologies Covered by @Scale Talks

Several write-ups describe infrastructure that routes traffic between services and delivers new code into production. ServiceRouter is Meta's service-mesh solution, in production since 2012, routing tens of billions of requests per second across millions of hosts and tens of thousands of services, with decentralised service discovery, load balancing and cross-region routing. Conveyor is the unified continuous deployment platform, whose pipelines pass each version of a service's code through nodes such as Build, Canary and Push, and which supports over 100,000 deployments each week. Owl addresses the distribution of large, hot content — executables, code artefacts, AI models and search indexes — within Meta's private cloud, using peer libraries linked into client binaries and trackers that manage the control plane.

• ServiceRouter — hyperscale service mesh integrated with the Thrift RPC transport
• Conveyor — continuous deployment platform with Build, Canary and Push nodes
• Owl — content distribution for objects ranging from about 1 MB to a few TB

Talks in the Systems @Scale track cover the data stores behind social graph reads and writes. TAO is a write-through cache on top of a sharded database that exposes an object and association data model, and historically provided read-your-writes consistency by routing a user's reads and writes to the same cluster. FlightTracker is a set of APIs and services that manage consistency across the social graph, providing uniform read-your-writes guarantees for both TAO and global secondary indexes while operating independently of the data stores. On the storage side, Everstore stores uploaded content as binary large objects, and Semantic Replication reduces the number of copies of video encodings that must be retained, saving over 30% of video storage.

• TAO — object and association store with simple query APIs
• FlightTracker — read-your-writes consistency across TAO and secondary indexes
• Everstore — BLOB storage system for uploaded content
• Semantic Replication — video storage optimisation saving over 30% of storage

Other entries describe cross-platform systems used in mobile product development. MobileConfig is a cross-platform configuration-management system, in production since 2015, that defines configs as sets of strongly typed parameters and pulls fresh values from the server at application start and periodically during a session; it connects to feature-flagging and experimentation backends such as Gatekeeper and Quick Experiment. msys, initiated by Messenger engineers in 2018, is a foundational messaging layer intended to be shared across the family of apps rather than duplicated per surface. The Design Systems Platform (DSP) provides a foundational layer for representing design-system elements as design tokens, and sits alongside theming infrastructure and a library of "chameleon components" that support Meta's more than 20 design systems.

• MobileConfig — strongly typed, cross-platform configuration management
• msys — shared mobile messaging layer
• Design Systems Platform (DSP) — design tokens synced into code and design tools

Product- and media-focused write-ups round out the catalogue. Threads is an app built by the Instagram team for sharing text updates and joining public conversations; official iOS development began in early 2023 and the app was released on 5 July after five months of technical work, with a read-only web version shipping a month later. On the media side, the site documents the adoption of the AV1 video codec, standardised by the Alliance for Open Media in 2018, for real-time communications, reporting around 30% BD-BR savings against an existing H.264-based solution. A companion post describes virtual video files, FFUSE and FUSE file systems used to process video uploads across Facebook, Instagram, Messenger, WhatsApp and Oculus.

• Threads — iOS app built on the Instagram codebase with a separate app delegate
• AV1 — codec adopted for real-time communications, with about 30% BD-BR savings
• Virtual video files — MP4-based abstraction connected to FFmpeg via FUSE