I felt the tension in the morning, saw the developer revolutionaries in the street, and in the first talk of the afternoon a shot was fired by the Jonathan Heiliger CTO of Facebook:
"I don’t know why they don’t, we’ve tried to communicate with them, but the system vendor OEM’s don’t get what it takes to build a server for Facebook."
[…]
"They should read the whitepaper Google wrote on it, they know how to do it."
[…]
"We aren’t seeing any of the performance gains in new micro-architectures that are being touted in the press. We just don’t see big gains."
Facebook is a daringly vocal company, with a CEO who once wrote "CEO…bitch!" on his business card back in the day, but the RFP’s they have been floating for their data-center build outs are very real. I’ve worked on responding to several of them and over a few years spend they are solidly eight figures, and high tens of thousands of servers. Hardly an outlier the infrastructure needs of Facebook are the needs of the coming cloud.
Their dissatisfaction with all of the system provider OEMs is a first shot in a coming war over data-center equipment relevancy. Google building their own server was seen as an eccentric oddity enabled by their unique profitability; if more and more scaled companies dismiss the offerings of the major OEMs, (and the performance gains of chips built with desktops in mind) it would begin to reshuffle what is now a $100B++ infrastructure market.
My angle: although I give the majors plenty of critique don’t count them out yet. Server designs are still rooted in the era of deploying one at a time; designs will change radically.
[...] Just say no to servers: Servers were essentially mentioned twice the whole day, other than in passing as a component of the cloud. “The biggest mistake we ever made was buying servers.” Matt Mullenweg, the founder of WordPress said, which was then followed in the afternoon by Facebook’s cajoling of sytems OEMs. [...]