Cloud Scale Erlang
summary
Title claims cloud scale needs ~4000 servers and an agent model is built using RabbitMQ Mnesia and Yaws to sync browsers phones and tablets; it runs but hits Erlang global operations and placement limits so clusters flatten at ~60–160 nodes unless partitioned. WhatsApp-style deployments push 7 million messages per second with 36 billion records and ~4 TB RAM across ~1,300 servers. Benchmarks show replacing TCP with RDMA gives ~33% higher message rate and global name/register ops drop scalability dramatically — need distributed supervisors placement and low‑chattiness primitives.
AI-written summary · about
chapters (8)
- 0:00Start
- 0:05About Myself
- 8:18The Memory Footprint
- 11:21Some of the Issues Faced
- 12:51Scalability
- 17:04The Media Streaming Service Used at Nasdaq
- 24:17Remote Direct Memory Access
- 25:56Networks and Scalability
related talks
26mIP Clusters - Scaling Zopim's Frontend Node Servers
40mDeploying node.js at scale
35mServerless WebSockets at Scale
16mHarnessing the power of the Erlang VM
34mBEAM: What Makes Erlang BEAM
01h 06mWhat every Node.js developer needs to know about Elixir
37mBuilding Fast Scalable Game Server in Node.JS (updated)
31mLessons from Scaling GraphQL to Half a Billion Requests per Minute