I saw a recent posting about Twitter moving some of their code from Rails to Java and it seemed to be germane to this conversation. It is a perfect example of the kinds of things you have to do for the tiny, tiny number of sites in the world that really need to scale:
I think what they did makes good sense for their vanishingly unique use case, and I think anyone building a new site from scratch would be nuts to try to build an architecture like that to knock out a site and see whether anyone was going to use it.
Rails is a great framework for building almost every web app
In that post, note that the performance improvement is not necessarily
a function of the language. My interpretation is that the major impact
comes from switching to a new async architecture which has been in
addition designed *a posteriori*, when you know where it hurts, and you
have the numbers, and you have a concrete technology ecosystem in
your company to evolve.
Agreed. It should also be noted that they only switched out the search piece that they had home grown wuth Ruby and were running in Rails for a search architecture that would work better. They looked across their entire stack and scaled/optimized the portion that was inefficient.