The fine folks at High Scalability have done a review of WhatsApp's talk at ErlangFactory, entitled High Scalability - High Scalability - How WhatsApp Grew to Nearly 500 Million Users, 11,000 cores, and 70 Million Messages a Second.
There are a ton of nice bits of knowledge in there, if you're contemplating scaling up a BEAM-based service to, I don't know, 50 million users.
- They use pg2, which means I should feel fine using it.
- They say to have no more than 8 processes accessing a given ets table to avoid contention.
- When they did have an outage (for 210 minutes), it was due to an issue with pg2's message queues.
Anyway, tons of useful stuff in there if you wanted to dig in further :)