06:55 PM
The Cost of Computing: The Internal Combustion Mainframe
For a few moments, don't even think about information technology. Rather, think about the automobile. There are roughly 750 million to 900 million of them on our planet. And even though today we have hybrids and electric cars, the internal combustion engine still powers 99 percent of them.
The internal combustion engine has been with us for more than 100 years. And while there are major issues with its use of oil and the impact of its emissions on the environment and climate change, the internal combustion engine has evolved and been refined based on a century of experience.
I recently attended the U.S. introduction of the Ferrari 458 in Greenwich, Conn. An integral part of this automotive masterpiece of art and design is a 275-cubic-inch internal combustion engine that produces 580 horsepower, or about 2.11 horsepower per cubic inch. It was not quite my father's 1970 Oldsmobile, which had a 350-cubic-inch engine that produced 125 horsepower, or .24 horsepower per cubic inch (and likely at fewer miles per gallon than the Ferrari).
Perhaps Moore's Law works for engines, too. In 40 years we have seen almost a tenfold improvement in efficiency. And just as with combustion engines, it is really hard to discount and toss away all that improvement in the name of replacing all computational engines with the "flavor of the day" -- servers, blades and perhaps even the cloud.
Now, please don't assume at this point that this article is about mainframe superiority. Rather, it is about the economics of computational platforms and the criticality of transparency into this dimension of technology economics in maximizing the value of IT to your business.
Platforms Not by Design
Most businesses today rely on an un-engineered mix of computational platforms: mainframe computers; midrange devices (e.g., AS/400); and UNIX, Wintel and Linux servers. The choice of how much to put on the mainframe or servers just sort of happens; it is not by design. At the same time, many organizations have adopted standards and architectures that move them away from mainframe computing under the banner of modernization.
While mainframe computing (or server-based computing) may not be right for all forms of computation, however, it is essential for organizations to consider both the economic considerations as well as the functional characteristics of their computing needs. For example, in the securities industry, the average IT cost per trade is, let's say, $0.17. Of the $.17, about $.04 is the computing cost. What if the right computing platform from an economic standpoint dropped that $.04 to .025? Consider 800 billion trades on the New York Stock Exchange annually, and a savings of a penny and a half a trade -- well, that is real money!
An analysis of data from 21 sectors (inclusive of governments) and 133 companies across those sectors reveals that the average company has a computing capacity of about .37 million instructions per second (MIPS) and about .17 servers per $1 million of revenue, meaning the average $10 billion company would have a core platform of 3,700 MIPS and 1,700 physical servers. (Obviously, as virtualization takes hold, this will change -- sort of like the impact of the introduction of fuel-injection on the internal combustion engine.)