Network World has an article about how Google builds its own servers that caught my attention.
Essentially, the article talks about how Google, in its desire to achieve the best possible value on the hardware – as well as to maximize power efficiency, actually builds its own hardware servers for use in its data centers.
What struck me most was the explanation on why Google actually uses inferior hardware obtained at a lower price-point that was put forth by Google’s senior vice president of operations, Mr. Urs Holzle. He explains that Google doesn’t actually need very reliable servers because it has written its software to compensate for hardware outages.
Now, I recently signed off on the purchase of two HP servers to the tune of $10,000 each. They are good, solid machines that have been running 24/7 for almost a month now without a single hiccup. However, the fact is that with some trade-offs on the specifications, I can actually build a “white box” – or unbranded server, for about a quarter of the ten grand price tag of the HP.
The question, on the other hand, is really about whether one is prepared to go with the slightly lower reliability of a white box. The traditional answer to that is a no-brainer: get the best hardware that your budget allows for.
However, I believe that we are in the midst of yet another paradigm shift in the way we consume IT services.
Just think for a moment about some of the most-used applications of all time. For Web mail, we have the likes of Gmail and Hotmail. For instant massaging clients, we have the usual suspects such as MSN Messenger, Yahoo Messenger and Jabber, just to name a few. Or how about the current king of hosted CRM, Salesforce.com, as well as the myriads of Web-based company portals or intranet applications in use.
From the above list, you will notice that the common denominator is that each of the mentioned applications are distributed in nature.
There are many more examples that I can come up with. Now, I certainly don’t mean to say that the days of high-reliability hardware is over.
Yet, suffice to say that I believe the absolute importance of the reliability of a single hardware box is ceasing to be as important as it once was. What is your take on that? Do you agree?