The Long Sad History of Microservices
summary
Argues microservices are mostly rebranded SOA and older ideas—actors, Erlang, VMS mailboxes—and shows trade-offs with concrete examples: a pricing service with 10,000-line "micro" service (C + CUDA), a rough throughput tipping point of ~5,000 TPS where distribution is justified, and the pattern of using event-driven async backends instead of synchronous HTTP between services. Covers DB-per-service vs shared DB trade-offs, container/process isolation spectrum, costs of continuous delivery, and practical security via encrypted events with key lookup and logging.
AI-written summary · about
chapters (19)
- 0:00Conference opening
- 9:44Microservices introduction
- 12:15Defining microservices
- 15:26History of SOA
- 17:13Early distributed systems
- 20:48Erlang and Smalltalk
- 23:44Why microservices fail
- 24:27First law of distributed computing
- 26:34Isolation in systems
- 30:52Lines of code myth
- 33:19Continuous deployment
- 35:06Database per service
- 38:44Containers history
- 40:28Synchronous vs asynchronous
- 42:36Autonomy vs authority
- 46:27Microservices overkill
- 47:31SOA fad comparison
- 49:19Academic foundations
- 53:10Trade-offs in architecture
related talks
38mMicroservices, a bittersweet symphony
23mWhy You Had a Bad Time With Microservices
32mMicroservices: A retrospective
44mMicroservices, Where Did It All Go Wrong
49mChallenges in Implementing Microservices
48mChallenges in Implementing MicroServices
44mIs It Time Already? Microservices Are Dead; Long Live Microservices!
24mDistant Past of Microservices