Meta’s AI Storage Blueprint at Scale

Summary

Meta describes how its BLOB-storage architecture evolved to address AI workloads' storage bottlenecks, which can stall GPUs and slow research iteration as datasets and GPU deployments expand. Legacy requests traversed multiple stateful metadata layers, sometimes across regions, producing latencies of hundreds of milliseconds even when flash made millisecond access possible. The redesign consolidates metadata into a flat schema backed by ZippyDB for O(1) path resolution, removes the dataplane proxy in favor of a fat client SDK, and deploys regional BLOB-storage stacks beside GPUs. For data loading, Meta combines host memory and flash with regional flash and global HDD-backed storage as tiered caches, using ordinary and deep prefetch, on-demand hydration, and TTL or LRU eviction policies. Production rollout saw rapid adoption, while both data-loading paradigms remain supported.

Takeaways

1. 1

   The redesigned metadata subsystem maps paths to storage addresses with O(1) lookup per chunk, replacing lookups across the namelayer, volumeslayer, and containerlayer.

2. 2

   The fat client SDK streams bytes directly from Tectonic instead of using a dataplane proxy, supporting the goal of zero overhead on top of Tectonic while staying within the power budget.

3. 3

   The tiered data-loading model uses host memory, host flash, regional flash, and global HDD-backed storage, with prefetching and capacity- or quota-aware TTL and LRU eviction.