Charles - I agree with Kevin, you can't believe everything you read on IBM's web site. I was involved with a customer who bought in on the idea of server consolidation in 2000. This was not the kind of overall consolidation that IBM is touting (which just incidentally requires a big investment in IBM GS), but a much more straightforward project - taking a number of operations running on Solaris, and on a number of smaller Sun servers, and putting them together on a big box (in this case a UE10000).
What the customer observed was that performance went down, and actual cost of capital equipment and administration went up. The UE10000 did pretty much what it was supposed to do - adjust partitions to manage load, etc. - but the benefits were not sufficient to outweigh the additional overhead of managing the system, and in addition the cost to get a transaction done on the bigger box was higher (an easy thing to see if you just run the numbers, mostly due to competitive price pressure on the smaller boxes).
Volume server vendors like the former CPQ also pushed this as a way to sell their larger systems, which in the Intel space were 8-way machines. They met similar economic challenges. An 8-way server cannot really replace 4 2-way servers, and the bus and I/O don't scale with the rest of the box. And with the base price of a 2-way dropping to commodity prices, the premium for a large system put the final nail in the coffin.
The trend today is toward loosely coupled integration, which exploits both the economics and modularity of contained, componentized smaller systems, rather than consolidation. |
