Agree 100% with most of what you've said, but I know this bit to be false: > ...you can normally get a pretty good feel for how your site will do under a load

Agree 100% with most of what you've said, but I know this bit to be false: > ...you can normally get a pretty good feel for how your site will do under a load using "ab" (apache benchmark) from your local machine. A single web server can easily out-serve the vast majority of internet connections you get at a home or office. I have 35 mbps down, 5 mbps up, and I can saturate my link (usually uplink) well before our web/app servers choke up. If you're testing a VPS at a provider like DO, you can spin up the cheapest Droplet in the same datacenter, then use it to test. Run your benchmarks in one shell session, while keeping another open with iptraf running to make sure you're not running in to network overhead. I know this sounds like overkill, and you qualified with "pretty good feel", which could very well mean "find out if your web server falls over at 10 concurrency", but it's always amazing to me how quickly bottlenecks creep in and muck up benchmarking results.