I read an interesting post on The Register that described how large service providers such as Amazon, eBay, Google, Microsoft and Yahoo! are choosing different options when it comes to fulfilling their server hardware needs in their ever-expanding data centers, while still trying to make the most of their data center resources (capital funding, electrical/cooling capacity, physical space, etc.).
Google (NASDAQ: GOOG) has opted to build their own server hardware instead of buying from a Tier 1 vendor like Dell, HP, IBM or Sun Microsystems. According to the post "Google builds its own servers. It also owns its own fiber, and builds its own switches. Most of you knew that."
I didn’t know this, but it does make sense considering the scale of Google's server infrastructure and the need to achieve some cost savings inline with that scale. I can understand building their own servers, but building their own switches sounds a bit bizarre to me. I mean are they actually buying a bare switch chassis, all of the internal components and then whipping out the soldering iron and going to town?
"Beyond crafting custom gear, Google tends to pick up cheaper parts than rivals. It uses lower power chips, cheap storage and lower-grade memory that some companies would never touch."
My first reaction at this was a bit of disbelief as I've always operated on that old "you get what you pay for" premise and avoiding downtime through component reliability/redundancy is a top priority for me, but again knowing Google's scale and the fact that their content is distributed among server farms, they do have the necessary redundancy in place at the server level, so there is logic to this approach.I've heard of some companies and customers opting to build their own desktops to achieve some cost savings and doing the same for a server isn't that much more difficult. Do you think this is an economical approach? I always take the Tier 1 vendor's reported MTBF (Mean Time Between Failures) with a grain of salt and have gotten new equipment that was DOA several times. Would you be more comfortable with hardware that your own personnel have built? What about the inherent support issue; are you willing to support the equipment you've built & deployed?
» Large Data Center Servers, Part 2 – Rackable Systems from ITechTips
This is a follow-on post to Large Data Center Servers, Part 1 – Google’s Whitebox Approach. I had described in the previous post how Google had opted to build their own servers, but also read in the post on The... [Read More]
Tracked on: December 28, 2007 9:26 PM | Permalink to Trackback