I came across this post on twitter’s blog. Really? Your all-time record tweets per second is 3283, and your systems are consistently over capacity?
I got the failwhale page several times in the past week. Seems to me to be pretty #fail that you can’t handle that relatively small of a load. This is the age of the internet, of cloud, of decoupling and taking advantage of radically different architecture concepts to scale beyond what traditional big iron could ever hope for.
That’s a patently absurdly low rate of small messages per second to handle, particularly when the medium allows for eventual consistency. Shame on you guys. You have enough cash, go get some guy who used to run systems for financial firms. Then get him drunk, and put him up in front of a whiteboard after getting him used to the idea that this data isn’t as important as trades, he doesn’t have to worry about milliseconds before a message that pops in somewhere pops out somewhere else.
Azul Systems is releasing a virtual appliance version of their JVM as part of their Zing Elastic Software Platform, so you don’t even have to buy a hardware Vega-based appliance. Maybe you should look into it.