Splitting the Monolith
summary
Meta converts its single monolith to a dedicated generative AI tenant that keeps the same developer surface but changes the infra: a smaller colocated deployment with 600 million lines of Hack, a longer request timeout, and about 10x larger thread pools plus JIT-cache warmup and shadow traffic. GenAI requests spend 90% of their time blocked on I/O, so these changes cut latency, enable streaming, remove timeout roadblocks, and make products roughly 30% faster in practice.
AI-written summary · about
chapters (5)
related talks
32mQuietly Strangling the Monolith - a real-world migration of a high traffic monolith
34mIt Used to Be Crazy to Migrate Monoliths. Now an Agent Does It.
43m"Monoliths are Simpler" and other lies we tell ourselves
36mThe Art of Decomposing Monoliths
25mOpening Keynote
50mSplitting the Monolith at Leroy Merlin
02mMark Richards at DDD Europe
33mChopping the monolith