Windows + Itanium: Mainframe killer? –

For a long time, Microsoft has been trying to invent all of the pieces necessary to cover the entire board in a continually changing chess game called the technology market. Consider the fact that the company plays in both the PDA and game/Webtop arenas, as well as maintaining traditional mainstream strengths in desktop and server operating systems and applications.

One of the places Microsoft desperately wants to play is in the NOC and WOC (network operations center and Web operations center) -- and it wants to put competitors such as Sun and IBM into the services business. To check its competitors, Microsoft needs the right silicon.

In the microprocessor game, there are has-beens, wanna-bes, coulda-beens, Great Silicon Hopes, and success stories. Intel's 64-bit Itanium processor, due to go into production any day now, could be Microsoft's hope at the high end.

A little history, or, sinking partnerships

Microsoft is having trouble at the top end of computing, where mainframes, hefty minis, clusters, and multi-CPU machines dominate. It's a nice place if you can get there, and it could fuel heavy sales for Microsoft if it weren't for a Rodney Dangerfield problem that's dogged Microsoft since its LAN Manager days. A quick refresher course in how to tie your wrists and ankles together is in order.

When NT 3.51, and then NT 4, were announced, several vendors came forth with processor families to support them. Digital (then Compaq) would run 3.51/4.0 on the Alpha CPU, Intel would proffer the Pentium, and back somewhere in the ozone were MIPS and the often maligned PowerPC chip from IBM/Motorola.

Essentially, forget all of them except Intel. Yes, Alpha development survives, but business plans for the rest of these processors running Windows NT/2000 are dead. There won't be any.

They're not endemic, but failed partnerships do mark the history of Microsoft. And they're that much more visible when they crash and burn like these CPU partnerships did. Each CPU maker except Intel put their chips into massive (read "expensive") workstation and server platforms that sold about as well as mud to a Floridian. While quite capable, these platforms cost megabucks in an era when Cyrix and AMD were attempting to cut the bottom out of the PC market. What were supposed to be eclectic, cross-platform successes became quiet, sad failures.

That was a shame. These high-end (for the time) CPUs that were to fuel high-end NT platforms were totally undercut by Intel and Intel-like processors. Very few vendors wanted to port applications from Unix or other operating systems to a non-Intel-family NT platform. Even Microsoft's own products weren't often or easily ported to the CPU-partners' platforms.

At best, these platforms lagged three to six months behind in development. Add market mismatch, changes in application structure, and the unwillingness of Unix (and other) developers to port to the new platforms. Then couple that mess to prices that were designed to recoup investment costs and mighty few being sold, and you get Silicon Valley landfill fodder.

The Itanium bullet: Intel's and HP's IA-64

Intel announced development of a 64-bit chip quite some time ago as the logical progression from the Pentium family. The chip went by several different code names, generally drawing on Oregonian geography. Meanwhile, Hewlett-Packard was in the midst of upgrading its PA-RISC family. In what will certainly prove to be a wise move, HP and Intel joined efforts to produce the IA-64.

Finally, after several years of development, trials, and missteps, Intel is about to bring the IA-64 into production. Itanium is designed as a weapon -- a bullet Microsoft and other OS vendors are already aiming at Sun and IBM.

Right now, the Pentium family is largely crippled compared to CPUs offered by both of these two companies. In particular, memory model restrictions prevent Windows from both running huge databases and crunching the data used by Web transaction and indexing engines -- applications that are becoming increasingly popular. The Itanium removes the Pentium model's 32-bit constraints, while providing what will be crucial competitive CPU horsepower.

Although Microsoft announced support for IA-64 long ago, it did so with no established timetable. Now, if you want to believe the story, an initial port will take place this year, with a more strongly integrated port finalized likely in 12 months. Single-processor machines will be able to utilize the CPU at that point. Support for multi-CPU machines will take longer.

The DataCenter edition of Windows 2000 server is designed to take advantage of up to 32 Intel Pentium-family CPUs per machine. (Deployments of the maximum number will likely be rare. Instead, 4, 8, 12, and 16 CPU platforms will probably be popular.)

But don't expect to take DataCenter and run it on anything but approved platforms -- the "Gold Hardware Compatibility List," or "Gold HCL." There's a reason for this: Microsoft has to achieve a fairly intimate relationship with a hardware vendor to guarantee that the combination hardware and software will work. Don't look for a wide variety of platforms for DataCenter, as it will take work on a hardware vendor's part to support ongoing development for it.

Is Itanium the fuel?

So does Itanium signal the death of current high-end Pentium platforms? Or will the marketplace become distracted, torn between the various combinations of clustered/SMP platforms supported by DataCenter versus the perceived musculature of the new chip?

Along with some shakedown problems, Itanium does offer the potential for a lot of improvement.

If you've been watching the memory models associated with the Pentium family, then you'll know that Microsoft has been trying to expand the Windows 2000 memory map from the NT model. The idea is to allow increasingly huge direct memory access -- beyond a maximum partitioned 4 GB of memory -- for applications and data. NT is good for 2 GB, or 4 GB with limitations. A 64-bit CPU can go much further. This increased direct memory-mapping capacity is what's needed by several popular applications -- notably relational databases and especially online analytical processing.

A 64-bit CPU also has the ability to potentially crunch twice as much information per CPU cycle. Coupled together, Windows 2000's expanded DMA model and Itanium's 64-bit capabilities will deliver faster and larger memory access and twice the throughput. Add these factors to Itanium's improvements in two very important areas, which are limited in current symmetrical multiprocessor models: memory and CPU cache management and coherency. The result will be Itanium-based SMP systems -- way-fast and way-cheap -- that are targeted at the heart of multi-CPU Sun systems.

MS competition and dependencies

Microsoft isn't the only vendor looking to anchor a future with Itanium. The Linux Trillium project, for example, has already completed a successful port to the IA-64.

The first place the IA-64 is likely to play is in the Web operations center, where developers want to add muscle to existing platforms or to collapse dedicated server groups. Here, only platforms that allow Web databases, indexing engines, and transaction platforms to scale will win respect.

Competition aside, there are key dependencies to the success of the Itanium/Windows 2000 combination. The first is a basic one that Intel has flubbed before: processor validation. There can't be any Itanium jokes about how many Itanium engineers it takes to change a light bulb (in Pentium terms, it's 2.999993938). Then there's the task that dogs Microsoft: porting an unbelievably complex code base to a new platform that will be rife with mysteries.

Inevitably, Microsoft and its hardware-platform partners who use Itanium will generate something they seem to like: upgrade revenues and a clear reason for mass migration on the high end of its BackOffice platform members.

Insider: How the basic tech behind the Internet works
Join the discussion
Be the first to comment on this article. Our Commenting Policies