They had thousands of machines, but it's hard to imagine why that is. It would be very interesting to get a description of what their hardware is actually doing

They had thousands of machines, but it's hard to imagine why that is. It would be very interesting to get a description of what their hardware is actually doing. Considering how StackeOverflow ran on a small handful of machines to a pretty large scale (and still isn't excessive), I can't see how more than a few machines should be necessary to run a site that serves small texts, even among millions of users. There is one caveat though and that's media. If media is transcoded, and/or if more than a small fraction of messages contain video - then the service might face using most of its resources towards that. But in that case , if I was running their ops I'd just disable media attachments while figuring out how to run the full service.