In my April 10th column about competing blade architectures I took a stab at predicting IBM's long anticipated blade strategy.
I wrote, "IBM is being fairly tight-lipped about the particulars [of its forthcoming "high-function" blade servers], but it's not difficult to speculate that it could offer solutions based on Intel or IBM processors, or both. If IBM marketed a blade with a single CPU using its Power4 shared-core dual processor ... On the Intel side, IBM has already been working with Pentium 4 Xeons in a variety of ways and, while they're not currently low-power processors, there's ample evidence that more than a few of you out there prefer punch over power savings."
So ten minutes ago, after an IBM phone briefing on its new eServer BladeCenter, I was struck by an odd sense of déjà vu. Yes, IBM's blades will initially be populated by Intel processors--Pentium 4 Xeons, Pentium IIIs need not apply--and Itanium and IBM's own Power processors show up next year. How is that, you ask? Aside from small footprints, blades are known for their low voltage, low power consumption characteristics. They're the Greenpeace of server platforms. Neither the Pentium 4 Xeons nor the Power CPUs make that grade. Not a problem, says Tom Jarosh, IBM's new blade VP, "Our customers want blades to be real servers and they want them to be ready not just for the edge of the network applications but all the way back into the enterprise." Processing power over electrical power, what a novel idea. IBM can back that up because it did an almost unheard of thing: It asked its customers what they wanted. "Their concerns focus around three factors: performance, density, and durability."
Having first taken care of performance with the high-end CPU menu, IBM described its blade density as being 2x 1U or 84 1U servers per rack. That's dense, but not to the earth-shattering degree that other blade designs facilitate. IBM says it was unwilling to compromise on performance or reliability in exchange for a higher density. And we can also suspect that by not using low-voltage CPUs, IBM would face many reliability issues in a very dense blade design. Do realize, however, that when the shared-core Power processors are officially added to the lineup next year, IBM's physical count of 84 servers will have the punch of 168 CPUs. (And IBM also hints at a 4-way SMP blade by next year.)
Durability has been addressed within the box through the use of redundant connectors, blowers, power supplies, switches, and so on that leave no single point of failure. Even the management module that handles the rest of the system is redundant in its implementation. On top of that there's IBM's Director Management layer with its own self-healing, self-provisioning, and self-management characteristics. It presents what Jarosh calls "a robust system that increases the confidence that customers have in deploying these technologies."
Durability also implies longevity, and IBM claims that its blades will have the legs to carry on for a while. You won't need to throw away the chassis to get the new blade model next year and the product is InfiniBand-ready, meaning that an InfiniBand switch can be snapped right into the BladeCenter. Connectivity with existing servers should be possible through IBM's Director Manager by running an agent on the server and managing it through a console on the blade.
Q302 is still a bit away and IBM's blades are just beginning their in-house final testing. While that usually entails some last-minute fine-tuning, expect to find both Windows and Linux (Red Hat and SuSE) available for them. The arrival of the Power processor next year opens the door to even more operating systems and that's when the fun should really begin. Don't expect the competition to shrug, pack up their wares, and walk away. Do expect to find a performance war, a price war, or both as the dueling gets hot and heavy.
What's your take on IBM's blade strategy? Speak your mind in our TalkBack forum, or send e-mail to Bill.
