On the Server Side: Alpha Users Are Chipped Off

Through a botched series of announcements, Compaq and Microsoft managed to alienate users who were loyal supporters. Here's what happened.

In August, Compaq pulled the plug on Windows NT for the fabled Alpha microprocessor. Through a botched series of announcements, both Compaq and Microsoft managed to alienate a series of users who were loyal supporters and performance-minded professionals. Here's what happened:

The 64-bit Alpha processor was introduced by Digital in 1993. Alpha's floating point processing performance has made it a real boon for scientific applications and its 64-bit and Very Large Memory Management capabilities made it a favorite for large scale database applications. Alpha supported Tru64 Unix, Windows NT, OpenVMS and Linux operating systems. In the past few years, the systems with the Alpha CPU and Tru64 Unix have made inroads at large e-commerce Web sites such as Amazon.com.

On August 20, 1999, shortly after the announcement of new Compaq 8-way ProLiant servers, Enrico Pesatori, senior vice president of Compaq's Enterprises Solutions and Services Group, released a memo to Compaq staffers. The memo announced that support for 32-bit Windows NT for Alpha would be discontinued. The memo specifically did not mention 64-bit Windows NT, not quite better known as Windows 2000. Nevertheless, many Compaq staffers believed that 64-bit development would continue.

At the time, that probably made a great deal of business sense: Compaq NT sales on Alpha were shrinking like a pair of bad jeans. From an estimated high of 15% in fiscal year 1997, NT share of Alpha server sales was approximately 2% in the fiscal year ending in 1999. These numbers gave proof that the longstanding strategy which Compaq inherited from the acquisition of Digital Equipment — provide NT on Alpha to drive volume shipments of the Alpha processor — was not working.

While this precipitous drop in shipments was taking place, the costs to support NT on Alpha remained. Compaq ran a facility known as DECWest in Bellevue, Washington near the Microsoft headquarters. DECWest employed approximately 100 engineers dedicated to porting NT code to Alpha and provided such essentials as BIOS code for various Alpha processors and Fortran and C++ compiler support. Analysts place the cost of operating the facility at approximately $1 million per year.

Given these two alarming numbers, the Compaq decision was not particularly surprising. What was surprising was the lack of planning and foresight. While the Pesatori memo said nothing about 64-bit NT on Alpha, it was widely believed that the 64-bit work would continue. This belief was based on the fact that 64-bit development for Windows 2000 at Microsoft was being done on Alpha systems. After all, Alpha was the only 64-bit system available. Most of Microsoft attempts to position NT as a worthy enterprise system centered on Alpha systems.

The much vaunted scalability events showing 1 terabyte MS SQL Server databases all ran on Alpha-based systems. Given the performance expectations of the first generation Intel/HP IA-64 CPUs (aka Merced aka Itanium), it seemed Alpha would be essential for Microsoft's high end Windows 2000 strategy. In fact, announcements in the middle of 1999 about 64-bit features such as the very large memory always said they would be first available on Alpha.

Within a few days of the Compaq announcement, Microsoft announced it would discontinue support for all Windows products on Alpha. So, next time you see Bill Gates ask him "whatever happened to cross-platform development?" Supporting multiple platforms was supposed tom enforce some discipline on software developers and make the OS code more high-level and less dependent on the quirks of individual CPUs.

When new processor technologies such as IA-64 were released, porting NT would be less of a problem because of this discipline. Remember when NT was first introduced? It was available for MIPS, PowerPC, Alpha and Intel processors. However, the development on processors other than Intel was always funded by — guess who? — the processor manufacturers. These manufacturers relied on the assumption of widespread NT popularity. When the sales never materialized, they could no longer justify the dollars. The supported processor list shrunk: first MIPS, then PowerPC and now Alpha.

I think Compaq thought Microsoft was somehow reliant on the Alpha for it's enterprise aspirations. But when was the last time Microsoft ever given any indication of its willingness to "play nice." But you can add that to the list of grievances the Justice Department has already noted. Getting Compaq to fund development for Alpha NT was a great strategy for Microsoft: it allowed them to grab the glory of high performance without the expense. And Compaq paid the price.

I find very little satisfaction in Compaq’s announcement that it will take a $50 to $100 million dollar charge in the fourth quarter of 1999 to pay for customer moves to new systems. Additionally, another $100 to $150 million charge will accrue in the first quarter of 2000. Microsoft, true to form, has been silent on the matter.

Of course, it's the users who get hurt. Compaq stumbled around for weeks after the Microsoft announcement trying to devise some strategy to placate its angry user base. It has finally agreed to offer users two options. If a user wants to stay with NT, they’ll be offered a trade-in on Intel based ProLiant systems. The value of the trade is tied to the age and speed of the Alpha system they're trading. If a user wants to stay on Alpha, they can get a license for Tru64 Unix or OpenVMS.

Neither solution is particularly satisfying because the value of NT and Alpha is performance. For users with large Exchange or SQL Server applications, Alpha was a great solution. While four and eight-way Intel servers can do the job, they are much more expensive than Alpha systems. Particularly hurt are Alpha workstation users who relied on great floating point performance for graphics and scientific calculation applications. Even though AMD Athlon processors outperform Intel Pentium III processors in overall benchmarks, floating point is not significantly better. Users of Lightwave and similar products simply have no place to go.

So, where’s the lesson in this tale?