Meta architecture: how Meta scaled (9 case studies)
The Meta write-ups we track cover TAO, Facebook's geo-distributed graph cache over sharded MySQL, which serves more than a billion reads per second; ZippyDB, a distributed key-value store on RocksDB with shard management and configurable consistency; the 2026 World Cup, when WhatsApp set a record of more than 30 million messages per second; and the hardware, networking and reliability work behind training large language models on very large GPU clusters.
We track 9 public engineering write-ups by Meta, published between 2013 and 2026. Techniques that come up more than once: distributed training (3), hardware reliability (2) and networking (2). Stack mentioned most: GPUs (2) and PyTorch (2). 6 of the 9 state concrete scale numbers, for example "20,000+ features per ads ranking request". Each card below links to our short summary and the original post.
Tech used in Meta's scaling write-ups
Meta scaling timeline: every story, summarized
- 2013-06-25
TAO: The power of the graph
TAO is Facebook's geo-distributed graph cache over sharded MySQL, serving the social graph at more than a billion reads per second.
- 2021-08-06
ZippyDB: Facebook's key value store
Meta's general-purpose distributed key-value store built on RocksDB, with shard management and configurable consistency.
- 2024-06-12
How Meta trains large language models at scale
Meta covers the hardware, networking and reliability work needed to train LLMs across very large GPU clusters.
- 2026-10-12
Shipping Media Over QUIC at Meta Scale for Instagram Live Ingest
Instagram Live ingest moved from Meta's in-house RUSH protocol to the open IETF Media over QUIC standard on iOS and Android. Edge endpoints translate MoQ frames into the RUSH format, so the backend did not change. Server-side routing let the experiment resize in minutes, the transport falls back to TCP when UDP is blocked, and the launch matched RUSH on quality and reliability.
- 2026-10-13
Meta's Generative Ads Model (GEM): Scaling the Central Model and Compute Engine for Frontier Ads AI
Meta replaced separate per-surface ad prediction systems with GEM, one central recommendation foundation model trained across all of its ad surfaces. A new LLM-style architecture (Kunlun) showed LLM-like scaling with 2.5x more model complexity, while custom low-precision kernels and 5D parallelism doubled end-to-end training efficiency to 20-25% MFU at 4x the training FLOPs across thousands of GPUs. Post-training transfer (LoopFM) and elastic pre-computation (Solaris) pass GEM's gains to real-time serving models without extra latency.
- 2026-10-15
2026 World Cup: Breaking Records, Spiky Charts and Reliability at Scale
How Meta absorbed the 2026 World Cup, when WhatsApp set a record of more than 30 million messages per second during the final. Pre-provisioned capacity buffers were the most effective lever, alongside country-level shedding of non-critical load, load-testing and resizing 100+ services, forecasting each region's peak separately, and no deploy freeze.
- 2026-10-19
Inside Prometheus: Meta's Journey to a Gigawatt-Scale Supercluster
Meta turned its New Albany, Ohio data center site from web traffic into Prometheus, a gigawatt-scale AI supercluster spanning multiple interconnected buildings that behaves as one machine. It runs NVL72 racks (Catalina on GB200 with 800G of scale-out per GPU, Clemente on GB300 in a single rack) on a dedicated Ethernet back-end network for GPU-to-GPU traffic. Reliability work pushed pre-training goodput above 85% and mean time between interruptions past 54K GPU hours, and the cluster is growing toward roughly 1 million GPUs.
- 2026-10-29
Catching Bugs Before They Land: JiTTesting's Human-on-the-Loop Approach at Full Scale
Meta's Catching Just-in-Time Tests (JiTTests) are now production infrastructure. Overnight, on spare capacity, the system picks risky diffs from the previous day and generates tests designed to fail: a "dodgy diff" workflow treats each change as a mutant, while an intent-aware workflow infers what the diff is meant to do, lists the risks, and builds mutants from them. An ensemble of LLM judges plus a deterministic rule-based assessor (RubFake) ranks the failing "weak catches", and engineers are only contacted for the strongest ones. 589 engineers have engaged, reporting 60% true positives, and in the first half of 2026 it helped catch 74 serious privacy bugs before landing, cutting privacy SEVs by 20%.
- 2026-11-03
F3 Lightning: How Meta Accelerates Ads ML Features to Production at Scale
Meta's Facebook Feature Framework (F3) compiles ads ranking features into one merged DAG so shared work runs once, but that graph was also the release unit, so one bad definition blocked every other feature as the catalog grew past 20,000 features per request. F3 Lightning makes each feature an independently compiled, validated, deployed, and rolled-back configuration, then rebuilds shared execution at service startup: stable identifiers derived from each computation's meaning, one flattened execution plan that deduplicates common nodes, a single per-request memory buffer, and dense integer slot mapping so request-path reads are direct array indexes. Comparable extraction features now reach production about 3.6x faster than traditional F3 with no increase in infrastructure compute cost.
FAQ
What is Facebook TAO?
TAO is Facebook's geo-distributed graph cache over sharded MySQL (with Memcache in its stack). Meta's 2013 post says it serves the social graph at more than a billion reads per second and millions of writes per second.
How did WhatsApp handle the 2026 World Cup?
Meta's 2026 post says pre-provisioned capacity buffers were the most effective lever, alongside country-level shedding of non-critical load, load-testing and resizing 100+ services, forecasting each region's peak separately, and no deploy freeze. WhatsApp set a record of more than 30 million messages per second during the final.
What databases does Meta use?
Two of the four Meta posts we track are about data stores: TAO caches the social graph over sharded MySQL, and ZippyDB is a general-purpose key-value store built on RocksDB and Paxos, with shard management and tunable consistency.