To thoroughly kill the joke, the idea is that you can write really fast software by writing your output to /dev/null (vs. disk) which basically makes write a NOP. Of course, it's completely useless in practice :)
web scale and sharding are just two other overused terms from the cited video (making fun of NoSQL fanboys).
I've seen this come up a few times, and benerved myself to ask the really stupid question: what is this for? Why am I supposed to want a process boundary between my HTTP stack and my applications, instead of running them inproc with mod_whatever? If it's a load balancer, how is it better than using a NAT to forward TCP streams and letting the app servers handle HTTP?
For one thing, mod_whatever may not exist for whatever hot new language you want to use. ZeroMQ already has bindings for most languages people might want to use (including weird ones) and there are only like 16 C functions plus some constant declarations, so bindings are easy to make. And once you have ZeroMQ support, Mongrel2 can talk to you. That may not be a killer feature for you, but it's really nice.
Plus, everything in the design of Mongrel2 just seems to make sense. I'm not sure how to describe it; Mongrel2 smells good.
This is looking pretty good so far. Anyone know if you can do a weight round-robin scheme with this? If I treat every server the same the smaller ones would get flooded in no time.
If you mean for the proxying, not yet. For the 0MQ handlers it definitely does, but some folks also want to do blocking before sending to the backends.
We've basically laid down the core features and then we'll start planning things like this. Page caching, different proxy backend schemes, filters, handlers, etc.
