Those of you who read my column regularly know that I believe in a resurgence of the need for big iron, or servers of mainframes and storage systems that fill rooms. The growth of mission-critical Internet applications - which must scale to support thousands of users a day or even your smallest customers and must check every order in the supply chain - almost assures this. And the emergence of the application service provider (ASP) market is further proof.
We already see this trend at work in the network switch business. Terabit routers are among the hottest-selling products, and more network equipment is now being sold to telecommunications companies and Internet service providers than to large corporations. So accept my premise for a moment that we'll need more big iron, then contemplate trends in the server market.
At International Data Corp.'s annual conference in March on IT market trends, my colleague Deb Goldfarb presented a 10-year server market forecast that sees a market overrun by entry-level systems and dominated by vendors that serve the consumer market and are trying to make a buck as prices keep falling. It's actually a very vibrant picture, one with new distribution channels emerging, new computer architectures and players, and 10 times more systems out there, albeit smaller systems. But the picture isn't very friendly to big iron.In the scenario most favorable to big iron, thedesigning and selling of big iron computersbecomes a niche market plied by specialty vendors. Sort of like today's supercomputer market.
The underlying trends behind this picture include the following:
-- Revenue from IA-64 processor-based systems passes that of RISC systems in 2004 (CISC is already pretty far out of the picture); a follow-on architecture emerges early in the decade.
-- A shift in the skill base and generational turnover mitigate "legacy" system inertia anddrive the success of outsourcing.
-- The customer base for servers shrinks as buying points consolidate, partly due to the influence of ISPs and ASPs.
-- Revenue from server shipments peaks at just under $100 billion in mid-decade before falling.
CIOs and others who must develop information systems to support this decade's new business models face this question: Will the new prevailing architectures (IA-64 and the follow-ons such as open-source software) provide the scalability, reliability, security and performance the new systems will need?
I don't see how they can. The idea of running the planet on Windows NT, on Intel-based servers, is scary. I'd have a lot more faith in systems like the IBM 3090. But the trend is turning away from monolithic, vertically integrated systems designed from the ground up to handle big loads. That's the big iron dilemma. Just as we need more of it, it's going away.
This may be a long-term picture, but as youmigrate to the Wintel architecture - and the numbers indicate most of you will - keep in mind the larger forces you're helping to unleash. The number of big iron vendors will diminish, as will the skill base and resources to design even bigger iron. In your own shops, the skills to install, maintain and program big iron will disappear. You'll have to rely on outsiders.
I suspect the easiest way to ensure system reliability and scalability will be creating redundant systems and designing to new peak-to-average-load ratios. That means you'll have to spend more money.
All this will happen over the next 10 years, right in the middle of many of your careers. While you focus on building those Internet-based applications and dealing with cataclysmic changes in the industry, keep a career-preserving eye on the big picture. Make sure you understand the long-term implications of this trend to swap out mainframes for farms of Wintel systems; make sure you understand how rapidly capacity demand can grow in an all-Internet world; and make sure you have Plan B in place for handling unforeseen processing and storage requirements.