This one is one of their less commendable efforts: yes, there was a discrepancy between what Heroku's docs said and the actual behavior of Heroku's routing. No, RapGenius's simulation is nowhere near correct: multi-tenant resource allocation is a pretty difficult stochastic optimization problem, and their simulation does not take this into account at all.

Heroku should own up to - and has owned up to, in my opinon- their communications fiasco. But I really, really hope the linked RapGenius article doesn't mislead folks to believe intelligent routing is easy (Tom Lehman, the then CTO of RapGenius, admits that he himself is a novice about distributed locks and such: http://s3.amazonaws.com/rapgenius/1361210233_heroku-adam-con...)

EDIT: minor grammatical clarifications

I did too. Learned a hell of a lot from them that's for sure, lol. Their Engineering team does a lot of, in my opinion, cool things and I really appreciate them taking the time to write about it.

The takedown on Heroku's dyno scaling was impressive. But also funny how they got themselves in that situation to begin with. If there's one thing about this world, RAP music will pay for $20k worth of Dyno's a month! Sad sad world.

edit - I like rap music.