May 13, 2011 | I guess the cloud is no longer technology’s darling. All the IT buzz now surrounds the Open Compute Project. If you are not familiar with the Open Compute Project take a quick look at http://opencompute.org/. You will see that this is really the brainchild of some bright engineers at Facebook, and the results are impressive.
Facebook makes it claim as running, “the most efficient (data center) in the world”. This revolutionary design not only saves massive amounts of energy but costs as well by removing centralized chillers and inline UPS systems. The Open Compute Project could very well transform how all new Data centers are built. I believe, it is only a matter of time before we hear captains of the industry trump energy and cost savings resulting from their organizations Open Compute architecture.
But is the Open Compute architecture really for everyone? The only way to properly answer this question is to closely examine the Open Compute Project. Let’s start with a review of the servers themselves. This is really easy considering Facebook provides the model number, technical specifications and detailed engineering drawings for each component (http://opencompute.org/servers/). For the most part the server components are straightforward. There is a chassis, motherboard, processor, memory, hard drive and power supply. All of these items are currently available commodities. Only the chassis, motherboard and power supply are not considered “off the shelf” components.
What is not so straightforward is the designed for failure mentality of these servers. Web 2.0 companies, such as Facebook, Amazon, and OpenX, are unique in their attitude towards reliability. These companies are looking to save costs by not requiring 99.999% reliability and can manage with 98% (or lower) reliability, think about this. These servers are designed to fail. Would you accept this failure level at your doctor’s office, or law firm or even your own place of business? The Answer is probably not, so why can Facebook and similar companies accept this lower reliability standard? Easy, they have all the money in the world to simply throw more hardware at the problem. If a machine fails, no problem there are racks and racks and racks of servers available. Most companies we work with do require 99.999% reliability or as close to it as possible. And certainly budget constraints are taken into consideration for every purchase.
Next let’s look beyond the server to the computing environment. For the Open Compute architecture to work you need to build a data center from scratch. Facebook kindly provides details for designing your own Open Compute data center http://opencompute.org/datacenters/). Just last week I was in Equinix’s Class A Virginia data center. This place was amazing, probably the best overall computing environment I have ever seen. But the Open Compute server architecture would simply not work here.
Why you ask? The answer is quite simple, the Open Compute server design relies on the outside air and an intricate misting system for cooling. Equinix does not provide this capability. If you put these Open Compute servers into Equinix they would burn up in about 1-2 days maximum. Because the outside air temperature is so critical for Open Compute to work, this will limit possible physical location of Open Compute data centers. Based on the information presented, I am not certain data centers in hot and humid environments will work. Facebook’s data center is in Oregon. I’m willing to bet the Facebook engineers immediately ruled out Arizona as a possible location.
Please don’t misinterpret the above as discrediting the Open Compute Project. I fully believe that this type of innovation is absolutely necessary. Our thirst for computing power is having a very real and negative effect on energy consumption and the environment. My organization (www.pssclabs.com) recently released our own revolutionary server design that promises to reduce carbon emissions by offering 55% recyclable metals (https://www.pssclabs.com/servers/1u-rack-mount/powerserve-duo-t2000/dual-server-technology). In addition, PSSC Labs (www.pssclabs.com) is working with component manufacturers to develop our own Open Compute server solution. I simply want to offering alternative discussion points before the marketing departments get wind of Open Compute and anoint it the next Information Technology catch phrase.
For me, this is the most interest discussion point not addressed by Facebook. Yes their data center may be the world’s most energy efficient but how does the higher server failure rate affect this measurement.
For more details about PSSC Labs’ high performance servers, supercharged workstations and industry leading clusters and clouds, visit us online at http://pssclabs.com.