Spolsky: Digg servers vs Stack Overflow servers

Stack Overflow is probably getting 90% of their page views through Google SERPs. That means nearly every request is a single answers page from a non-logged in user. Stack Overflow could write out static HTML files for 90% of their traffic. That would let them scale to &#62;500M page views on one server.<p>Digg on the other hand is generating highly dynamic pages on a per-user basis for a lot of their traffic. That's inherently more resource intensive.<p>That said, Digg clearly wasted a ton of money on servers. They pretty much had to. The only way to spend $40 million is to over hire and over build your datacenter.

Incidentally, we now serve about 24 million page views a month off 2 servers. (We serve images off www.)

I mentioned this to Spolsky already but depending how you want to count servers (web servers or all front facing servers or all servers in your infrastructure?), we have 10-20 servers servicing 350 million page views per month. That is of course not counting all the ajax requests that we also service, which don't count as page views. So we service 17-35 million page views per server, compared to StackExchange's 12 million page views per server.<p>We use PHP and Apache, so the gloating about ASP being way more efficient than PHP seems to be unfounded.

I handle over 50 million page views a month with a single server on a large (7.5GB RAM) EC2 instance. 2-3 MySQL queries per page view, no caching. It's hard to imagine not being able to cut Digg's infrastructure size by an order of magnitude with the right architecture and perhaps more RAM per box than whatever they're doing now.