Comments on: How Facebook Changed Technology in One Day

By: Jonathan Heiliger

Jonathan Heiliger — Mon, 11 Apr 2011 20:44:03 +0000

Hey Mohan… thanks for taking an interest in the Open Compute Project. As we said last week, we’re starting at v1.0 and are looking forward to collaborating with like-minded people on how we can collectively improve application, server and datacenter efficiencies. The opencompute.org site has already been updated with additional material and we are working on posting more detailed specifications as the partners provide permission to share their work.

By: Christoph Weber

Christoph Weber — Sat, 09 Apr 2011 19:08:33 +0000

I think you’re a bit harsh here, and so is the author of the linked blog. For example, the power supply in FB’s servers is innovative, as is their battery backup solution, and the design docs allow others to come up with competing implementations if they are sufficiently knowledgeable. And because the designs are GPL’d there is no road block whatsoever to reimplement anything.

Moreover, Facebook’s approach is a nice example of leaving nothing unquestioned when designing and implementing a new data center. Just sharing the thought process is already worth a lot.
I wish Google and Amazon were similarly open.

By: Christoph Weber

Christoph Weber — Sat, 09 Apr 2011 18:59:38 +0000

As a facility designer/maintainer you have no idea what the servers do and simply assume they’ll do something useful. Delivering on that assumption is the job of the programmers and sysadmins. The facility guys make sure that of the power entering the building as many Watts as possible are used to power servers, as opposed to being used for overhead, such as cooling, lighting, various losses, etc. In this context work simply means keeping the servers humming.

The servers convert those Watts into heat, of course. No useful mechanical work is done, and the information entropy is completely negligible.

By: Mohan

Mohan — Sat, 09 Apr 2011 16:16:02 +0000

Hi
Please read the following article before going gaga abt nothing here. I am really disappointed your not doing your research well before posting in a quality blog like OM. Very disappointed.
http://laforge.gnumonks.org/weblog/2011/04/09/#20110409-facebook_opencompute_hot_air

By: J Flash Rodriguez

J Flash Rodriguez — Sat, 09 Apr 2011 15:16:48 +0000

Powershifting since 1990 and always learning new things. With PUE being important to climate change, thank God for constant innovation. Having worked at the many fine technology institutions mentioned , PUE has come a long way in 10 yrs. Like Alvin Toeffler said.

By: Bob

Bob — Fri, 08 Apr 2011 06:44:50 +0000

What does it mean to “do work” in this context?
Deliver an end result of some sort? What would that be?
Aligning magnetic domains on a disk platter? Sending real info encoded as a burst of light onto a fiber that actually leaves the building?
Beyond those, everything else winds up as heat (ok, perhaps a stray RF noise emission might escape the building and planet and propagate through outer space :-)

By: Duncan

Duncan — Fri, 08 Apr 2011 06:03:41 +0000

To achieve a PUE of 1.0 or lower would require the laws of physics, as we know them, to be bent and broken. A PUE of 1.0 means the machine, as a whole, is 100% efficient – every watt delivered to the datacenter for machine use is used to do work, not lost as heat or in voltage conversion inefficiencies. A PUE under 1.0 would mean you have invented perpetual motion.

By: ChristophWeber

ChristophWeber — Fri, 08 Apr 2011 04:11:15 +0000

You can’t have a power usage effectiveness of 1.0, much less below 1. Your servers consume 1.0 units of energy by definition, and the difference from 1.0 to, say, 1.07 is how much you spend running the rest of the data center. Cooling, lighting, losses, etc.
FB’s 1.07 and Yahoo’s 1.08 is damn low and extremely hard to beat.

Myself, I am looking forward to a new data center in May running at a PUE of 1.2, which is still considered cutting edge if you don’t have the luxury of building from scratch like the big boys.