> It seems like a reasonable idea to try to evaluate a backbone ISP from a > traffic speed perspective, from a random sample of my customers (94% of > whom have dedicated, high-speed connections) to the nearest hub of the > ISP. I need no services from the ISP other than basic throughput, my web > server to customer. > > After spending a considerable amount of time at this, I've found the > decision generally seems to be made on a hunch. Large sites have > redundant service from two or more backbones, but I am not near that > stage. > > So, logic be damned, I guess it's time to flip a coin.
Take a look at ab, the Apache benchmarking program though, that's what the webperf site uses. They poll 4 times an hour, and try to use a number of different poll hosts, located at a number of main switch points. Rob
