THE WORLD NEEDS ONLY FIVE COMPUTERS
By Gregp on Nov 10, 2006
And, no, I'm not paraphrasing something that I bet Thomas J. Watson never uttered in 1943 anyway. But he should have because, ultimately, he might turn out to have been right.
Let's see, the Google grid is one. Microsoft's live.com is two. Yahoo!, Amazon.com, eBay, Salesforce.com are three, four, five and six. (Well, that's O(5) ;)) Of course there are many, many more service providers but they will almost all go the way of YouTube; they'll get eaten by one of the majors. And, I'm not placing any wagers that any of these six will be one of the Five Computers (nor that, per the above examples, they are all U.S. West Coast based --- I'll bet at least one, maybe the largest, will be the Great Computer of China).
I'm just saying that there will be, more or less, five hyperscale, pan-global broadband computing services giants. There will be lots of regional players, of course; mostly, they will exist to meet national needs. That is, the network computing services business will look a lot like the energy business: a half-dozen global giants, a few dozen national and/or regional concerns, followed by wildcatters and specialists.
Let me back up and explain what I mean by a Computer, and then why I think this is inevitable. I mean "Computer" as in the "The Network is the ...". These Computers will comprise millions of processing, storage and networking elements, globally distributed into critical-mass clusters (likely somewhere around 5,000 nodes each). My point in labeling them a Computer is that there will be some organization, a corporation or government, that will ultimately control the software run on and, important to my argument below, the capitalization and economics of the global system.
These Computers will be large for a number of reasons. It seems that the successful services are most definitely growing faster than Moore's Law. That is, in addition to upgrading to faster systems they are adding more of them and the compound growth is getting pretty spectacular in several cases. A company like Salesforce.com sees hypergrowth not in the form of some intrinsic demand on CRM (within an average company, definitely not growing close to Moore's Law --- Enterprise CRM is overserved by systems performance improvements), but rather the sum of consolidation of CRM systems across thousands and thousands of companies. Live.com is likely to fall into this camp, too. The growth seen by a Google or Yahoo!, on the other hand, is more directly a function of their pipe-filling roles: the greater the end-user bandwidth, the greater the demand on their infrastructure.
Moreover, there is most definitely an economy of scale in computing. To the extent that there is a scalable architectural pattern (cluster, pod, etc.), the per-unit engineering expense gets amortized over increasing capital volume. So, more and more engineering can be invested in driving higher and higher efficiencies at scale.
Our bet (meaning Sun's) is that, like the energy, transportation, telecommunications and power utility businesses, most of these companies will realize that they can become even more efficient if they rely upon a few, highly competitive and deeply technical infrastructure suppliers (think GE, Siemens, ABB for power systems, Boeing and Airbus for commercial aircraft, Ericsson, Nortel, Lucent/Alcatel, Nokia for telecom, etc.).
All this being said, a large enough enterprise (say, a big financial services firm) still have some pretty compelling reasons to build their own Computers. My only advice here is to approach the problem as one of latent scale. That is, think that you are building one of the world's five, but you just haven't quite grown into it yet! Same advice goes to start-ups: because either you will grow to become one of the big Computers, or you'll be acquired and be Borg-ed into one of them!
Naturally, we aim to be the premier infrastructure supplier to the world's Computers. Blackbox is just the beginning (More on Blackbox in a previous entry). Whatever its form (or color!) the emerging infrastructure will be far more efficient than what we think of for conventional enterprise computing. And, just as a reminder, that doesn't mean its piles and piles of cheap boxes, any more than you'd design a power plant with piles and piles of cheap portable generators. In the latter case, the little problems of noise, pollution, reliability and conversion efficiency are scaled into some really nasty ones.
Similarly, the cheapest computing is not necessarily obtained by lashing together racks and racks of the cheapest computers you can find. Engineering for scale matters. Really matters.