I stay quasi-anonymous on here, but I worked at a company that bursted -- at the time -- to approx 1.5m requests per second. Average traffic was approx 400k rp

I stay quasi-anonymous on here, but I worked at a company that bursted -- at the time -- to approx 1.5m requests per second. Average traffic was approx 400k rps. It was run on naked boxes in many pops worldwide. We heavily used the jvm, which dramatically simplifies application distribution. Boxes were all netboot; they came up, grabbed the OS, and based on some pretty simple config figured out what jars to download and run. We also costed out a move to aws, and it was somewhere between 5 and 6x as expensive as buying our own hardware. Thousands of boxes worldwide managed with a team of two fulltime sysadmins plus two more primarily engaged in dev with sysadmin duties as-needed, plus remote hands services in the various datacenters.