CONCEPT Cited by 1 source
Fat client SDK¶
Definition¶
A fat client SDK is a client library architecture that embeds substantial storage-layer capabilities — block fetching, caching, peer coordination, metadata caching — directly in the client, eliminating intermediate proxy services from the data path.
Trade-offs¶
| Benefit | Cost |
|---|---|
| Lower latency (no proxy hop) | More complexity in the client |
| Lower power footprint (no proxy servers) | Harder to upgrade uniformly |
| Higher throughput (direct streaming) | Client must handle failover, retries |
| Enables distributed peer caching | Version-skew risk across clients |
Instance at Meta¶
In Meta's BLOB-storage re-architecture, the fat client SDK:
- Embeds a Tectonic BlockClient for direct data streaming from storage servers
- Integrates Owl peer subsystem for distributed data caching across GPU hosts
- Includes a ReadPlan cache for 1–2 ms metadata access
- Implements dynamic concurrency control based on application-level congestion signals
- Exposes a deep prefetch API for background hydration
The result: zero overhead on top of Tectonic, with all caching and intelligence client-side.
(Source: sources/2026-07-01-meta-ai-storage-blueprint-at-scale)
Seen in¶
- sources/2026-07-01-meta-ai-storage-blueprint-at-scale — canonical instance for AI storage at Meta