COMMENTARY--When the IBM PC arrived, expansion slots didn't look that exciting but they changed the world of computing. Twenty years on, blades could have the same impact, by simplifying the mess that early PC architecture has led to.
When IBM jumped into the PC market, it didn't realize what it was doing, and accidentally developed a sensible strategy. It set up an open platform (such a departure for the IBM of the 1980s that no-one doubts it was an accident), and the rest of the industry jumped on board. The Intel architecture, boosted by the PC, was such a success that for a long time IBM struggled to keep up with it. The company has only seemed to catch up now Intel has become a serious server platform.
In blades, IBM's ideas look a bit more planned, but once again there is a big element of luck involved. And, this time, the company has competition right from the start, as others consciously try to make that architecture right, and spot the opportunity that will follow.
Blades represent a somewhat incestuous moment in IT history, as they address an opportunity for Intel processors that has been opened up by the very dominance of those Intel processors. IBM consultant Tikiri Wandaragula talks about different architectures evolving like different animal species.
Now, if I take this analogy further, it seems that the Intel processor could be seen as being like the human species, which would mean that that RISC chips are like apes, which still continue to exist in their niche environments, even though a more dominant form of life has appeared.
As David Attenborough showed so clearly in the "Life of Mammals", which recently aired in the UK, the human species is an opportunist and a generalist. It has spread chaotically, colonizing different parts of the world, making other species extinct and is seemingly in danger of choking its own existence.
All this is not unlike the Intel processor. They are everywhere now, forming the basis of at least two hostile cultures (Lintel and Wintel). Their expansion is completely out of control, as many IT managers would testify. While the ape-like RISC processors peacefully munch leaves in their jungle, and the whale-like mainframes plough steadily on in the uncharted deeps of the datacenter, Intel processors are ripping up whole areas of the IT infrastructure and turning it into an Intel-only habitat.
Blades look like bringing civilization to the uncontrolled barbarism of Intel, allowing IT managers to bring them under control. The idea is that the hardware will manage itself, instead of making demands on the administrators, and everything you need will fit into a well-managed, compact chassis.
As an aside, it is worth explaining how anything as "new" as this can be sold in a recession, when only the most conservative technology has any chance of survival. It's a "cost-of-ownership" sale, a consolidation sale. The message is that the chaos of Intel proliferation costs money, in terms of rented rack space, and in terms of administration time. Keeping track of all the servers you have, for Exchange, SAP and all the rest of it, is one of the biggest demands on an IT manager's time.
"If a project has a convincing cost-of-ownership study, then any delay costs you money," said Wandaragula. Delaying other projects can be seen as saving money. For a small fee, IBM will do you a cost-of-ownership study, of course.
In the long term, blade servers could even, finally, undo the piece of Intel proliferation that has caused IT managers the most stress over the last 20 years: the desktop PC. The words "thin client" are deeply unfashionable, but they are starting to be said with more and more conviction by IBM consultants wanting to sweep more and more IT complexity onto the shelves of their blade chassis.
Another image might be the move from free-range farming to efficient, intensive, factory farming.
Of course, blade servers can't do everything yet. So far, they can't have fast enough interconnects to let one server address the memory of another server as if it were local. The economics of blades is they are made of commodity parts -- though they now have full-blooded processors, not the mobile variants the first blades had, and they have redundant power supplies and other back-up equipment. Fast interconnects are not a commodity, so they only appear in proper servers, like IBM's x440.
The solution is to use blades for what you can -- all those dedicated Intel servers -- and use "proper" servers liked IBM's x440 for the applications which go beyond a single blade. Slide the "proper" servers into the same chassis, and you still tidy up a massive amount of the datacenter.
Get third parties like Nortel to make blades for other jobs like LAN switching, firewalls and VPNs and you tidy up still more of the corporate IT landscape.
All of which makes IBM's blade strategy look very good compared with some others. Sun's, for instance, focuses on SPARC processors, which haven't proliferated nearly so messily, and don't require consolidation so urgently. Though blades based on other processors than Intel do make sense, it seems far more likely that the net movement of applications will be from Unix systems to Linux on Intel, rather than the other direction.
So, safely encased in blades, Intel processors look set to make more inroads. In future, we can hope, they will be less of a parasite in the datacenter habitat, and more of a beneficial species.