Of morons and monoliths

Subject: Windows - the ultimate componentized monolith
During the Microsoft trial and during Bill Gates' testimony in March of this year he vehemently stated that any unbundling of Windows would destroy both Windows and Microsoft. During the same entire time he was saying this his company was fervently promoting products which did exactly that - Embedded Windows - which can run on as small a footprint as 200k (Embedded CE.net) or 5mb (Embedded XP). See here.

Here is Bill Gates in March, 2002, telling us of the perils of viewing Windows in any way which was not monolithic

c. Consequences of Code Removal 224. Even if it were feasible to develop operating systems in compliance with Section 1's design requirements, the resulting products would provide little value to the marketplace.

225. Section 1 contemplates a world in which OEMs and anyone else that licenses 10,000 copies of Windows can remove the binary code of any "Microsoft Middleware Product" in Windows, while the balance of the operating system will continue to perform effectively. In other words, Section 1 contemplates a world in which there is not one version of Windows, but many. These different versions of Windows would each expose APIs.

226. If Windows were deconstructed and offered by OEMs and others in multiple variations, the primary value that Windows provides to the marketplace—a stable, consistent platform for software development—would be lost. Neither ISVs nor end users would have any assurance as to what functionality would be provided in any given variant of Windows.

227. For ISVs, the advent of multiple versions of each release of Windows would make it impossible to design products that rely on the presence of particular functionality in the operating system. ISVs would thus be forced to include more functionality in their own products, making those products more time-consuming and expensive to develop and larger and more complicated generally, leading to more bugs and higher support costs. Many existing products would not run properly, or at all, on versions of Windows from which the binary code supporting necessary APIs had been removed. (See Demonstrative Exhibit 2.)

228. To provide a satisfactory customer experience, ISVs would have little choice but to develop new products designed to run only on specified new versions of Windows (e.g., the Compaq Presario 2002 version), returning the industry to the fractured state from which MS-DOS freed it in the early 1980s.

229. Consumers would likely be confused and frustrated if OEMs and others were allowed to remove key parts of Windows and nonetheless sell the resulting software as "Windows" (as Section 2.b of the NSPR would expressly permit). Imagine the disappointment that would result if a consumer purchased a new Windows-based PC, only to discover that key new features (such as video instant messaging or Internet support) were missing and various applications would not work at all because of missing APIs. OEMs would have strong incentives to fragment Windows in this way because they would be paid by Microsoft competitors simply to remove code from Windows, or to remove code and replace it with competitors' code. Of course our competitors could contribute to such fragmentation directly by licensing 10,000 copies of Windows, which they would modify themselves and distribute to end users. If OEMs and others were permitted to produce myriad versions of Windows with varying feature and API sets, and identify them using the Windows trademark in order to trade off the goodwill Microsoft has developed in that famous name, the Windows trademark would become meaningless.

230.If OEMs offered different versions of Windows, there would be (i) significantly less price competition among OEMs (because their products would be less interchangeable); (ii) greatly reduced interoperability between brands of PCs (because key interoperability technologies, such as protocols, might be different on different machines); (iii) greatly reduced interoperability between applications and PCs (due to missing APIs); (iv) increased development costs; (v) reduced development output and innovation; and (vi) considerable consumer confusion. These harms would likely lead to slower growth of the PC industry as a whole.

Here are Gates' men praising the versatility of Windows as an OS which can be componentized. Bill_L[MS]
A: We are not based on the Home Edition of XP. We are based on XP Pro, and we don't have a Home Edition component.
Jon_MS
A: Windows XP Embedded has the same system requirements as Windows XP Professional, although in certain situations, you can get by with less hardware (older processor, less RAM, etc).
John_MS
A: XP Embedded is an extremely customizable platform to build embedded run-times based only the components that you choose. This means that you can absolutely tune the OS to your specific device needs.
David_MS
A: The latest Windows technologies- h/w & s/w support, multimedia, connectivity, etc- are all present in XP Professional. XP Embedded enables you to pick & choose and implement any combination you wish into your embedded solution.
John_MS
A: The Windows XP Embedded platform is client only, for Server Applications please see other Microsoft offerings, like the Server Appliance Kit.

We can't keep the government, our laws and our judiciary from being manipulated and undermined, but we can at least be clear in our own minds about who is saying what, and whether or not the people doing the talking are very devious and cynical, or not.

Gene Mosher

Subject: DEC's Alpha created HP panic, desperation

I'm surprised you printed something so one-sided. It was obviously written by a former DEC customer who was under the silly impression that because the Alpha was on top of the SPEC charts most of the time, it would end up beating out PA-RISC, POWER, Sparc and IA-32. Now he's unhappy that Alpha's been killed in favor of Itanium. Perhaps he hasn't noticed the financial state of the computer industry, technical merit alone doesn't decide whether a product makes sense, it has to make money. And Alpha was always at least 4th in sales, behind Sparc, PA-RISC, and POWER. Its hard to spend the hundreds of millions new generations of chips require when you have to amortize it over such a small base. There were some people upset that Cray never released the Cray-4 and beyond, because nothing could touch it for supercomputer work. But if they only sold a handful of supercomputers a year, they couldn't make money at it, so now Cray's next big supercomputer will use AMD's Hammer. Welcome to the real world, dude.

There are several complete fabrications his letter that I'd like to correct, however. First off, HP designers had very little to do with Itanium 1, that was an internal Intel effort. Itanium 2 was mostly designed by HP, and reflects their design principles better -- HP always believed it was better to have a small team with many years of experience each instead of Intel's approach of an army of people barely out of college supervised by a few veterans. Intel is very good at designing x86 CPUs, but everything else they've done in-house has been a big flop (remember the i432 and i860....no? proves my point!) There was no Alpha technology stolen or used in Itanium 2, Intel didn't acquire that until just before Itanium 2's release! The first IA64 design that will include Alpha technology and ideas (and be worked on by the Alpha team Intel acquired last year) is roadmapped for release in 2005 or 2006. The design cycles are long for such a complex chip.

In addition, while no one disputes that Alpha was on top of the SPEC charts for most of its history, HP's PA-RISC chip typically tipped up on top for a few months each time a new rev was released. No mean feat, considering that it has been using essentially the same core for all of the PA-8xxx series, the original of which was released in 1995. Trying to squeeze greater performance out of their dated PA-RISC core started working less and less well in the past couple years, and yes it has since fallen well down the list. How well would Alpha have been doing if it was still using the core from the 21164 that was recently around the same time? Since HP's plan is to migrate to Itanium, that's not really a big issue. It wouldn't have made sense to HP to invest extra resources to do a new core for PA-RISC to make it competitive, they'd just have a harder time getting customers to migrate to Itanium. Anyway, while HP doesn't look all that good in the SPEC race these days, their systems still stand at or near the top of TPC charts, and that's the market HP is primarily interested in. People who wanted pure number crunching were already using DEC Alpha or SGI systems anyway.

I've included my name/address, but if you print this, please don't include them. Doug

Subject: DEC's Alpha created HP panic, desperation

Ye gods and little fishes! How many errors can one person make in a single letter? Are you running a competition on that?

Your article "Hewlett Packard developed the Intel Itanium 2" is accurate, to the best of my knowledge, incidentally, and asks some questions which I have asked myself (without getting an answer). You have answered part of it when you pointed out that very few of the Alpha people did actually go and work for Intel.

I can tell you that Intel stated explicitly back in 1994/5 that the Merced would be designed by Intel and the McKinley by HP, with at least one of the (original) Madison and Deerfield being designed by Intel. That was the plan. All the information I have heard since confirms that it was the actuality, too, though Intel took over the project from HP a few years back (just before Fiorina).

I have reason to believe that HP had taken a decision to cancel (or, at least, shelve) their Itanic plans on the grounds that the technical staff of HP were clued-up enough to admit their mistake. But then Fiorina came in and overrode them!

My best guess (largely confirmed by Email) that Intel sealed the fate of the original Madison and Deerfield during the period when it was cannibalising its own long-term development teams in a vain attempt to get the Merced working, performing and out. Fred Brooks could have told them why that could never have worked ...

But, for heaven's sake, that project fizzled out in 2000 at the LATEST. What on EARTH have those people been doing since?

Name and address supplied, withheld. µ