>I'm challenging you to provide the data that apparently is still missing that proves that this was not 'a localhost test'.<
I think what he was saying in relation to it being not 'a localhost test' is that it isn't a test that touches the network stack, even to the loopback interface -- you seem to be equating localhost with anything that happens on a single local machine regardless of whether it has anything to do with the network or not. A difference in use of terminology, perhaps.
Yes, nobody seems to get confounding at all. I'm testing poll vs. epoll on selecting active file descriptors to compare their performance. I'm not testing anything else, not claiming anything else. To then test that with a full on server that has an Amazon EC2 cluster blasting requests with HTTP parsing and serving files would completely confuse the analysis.
You don't measure a specific thing by inventing some full on "real world" evaluation. You test that specific thing. Why is it this simple concept seems to baffle so many coders?
Your results seem correct to me, even though I'm a bit disappointed at the actual number (I was hoping for 0.8 or 0.9, not 0.6), and the code looks good. I'm meaning to run it for myself, but didn't have the time to do so, I should get on it shortly.
I still of the opinion that a >0.6 ATR means you're fucked anyway (where processing all those fds can't be done as fast as data arrives), but I'm trying to come up with a good small test for that hypothesis (like you did).
Until I do, it's just an opinion/hypothesis, and it's kind of hard to argue about it strongly either way.
I think what he was saying in relation to it being not 'a localhost test' is that it isn't a test that touches the network stack, even to the loopback interface -- you seem to be equating localhost with anything that happens on a single local machine regardless of whether it has anything to do with the network or not. A difference in use of terminology, perhaps.