I read these two blog posts:

• https://blog.cloudflare.com/how-we-scaled-nginx-and-saved-the-world-54-years-every-day/
• https://blog.cloudflare.com/the-sad-state-of-linux-socket-balancing/

Let's be clear, I don't really know what I'm talking about here. But it sounds interesting, and I'd like to better understand where we are today, if some of these trade-offs and ideas would make sense for us, and to perhaps document limitations or restrictions for why some improvements would not make sense or are infeasible.

To do for @Krinkle (and possibly document the result of these and/or file follow-up tasks).

• Identify the services between text LVS and app servers.
• For each, do we use a shared or central accept queue? (e.g. can work relating to one slow response hold up other requests waiting for that worker's queue?) Or, does the question not make sense in our context?
• For each, would changing this make sense? Is it worth experimenting with the other?
• Does the consideration between blocking-accept and epoll-and-accept apply? If so, are we likely in the optimal state, or is there something worth experimenting?