Pacific Connection(英語)

For Intel, UNIX is Worth the Pursuit

It has been called the "Wintel" alliance. Just as Microsoft has earned riches by producing the default desktop operating system, so has Intel reaped handsome rewards by producing the most popular CPU to run Windows. Today, some 85 percent of all computers are powered by Intel technology, and the company has plowed millions of dollars into an ambitious corporate "branding" campaign to keep it that way.

Given this comfortable position, you might think that Intel would be reluctant to tamper with the marriage, especially given Bill Gates' reputation as a jealous partner. But when it comes to operating systems, the chip maker is keeping its options open. Over the past year or so, Intel has given financial and verbal support to three versions of UNIX, Windows NT's main competitor on the server, while acting as a catalyst in the development of middleware that would simplify the development of UNIX device drivers.

  • In September, Intel joined Red Hat Software, one of the leading distributors of Linux.
  • Intel has worked with IBM, Santa Cruz Operation (SCO) and Sequent in their effort to create a single UNIX version between them.
  • Last December (1997) Sun and Intel announced that Sun's 64-bit Solaris 7 operating system, initially destined for the SPARC architecture, will be optimized for Sun's IA-64 processor, which is now expected to ship in mid-2000.
  • Intel is playing a central role in the development of a uniform driver interface (UDI) specification. The interface will allow device drivers to be portable across both hardware and operating systems without any changes to the device source.

All of this activity serves to make Intel an active player in virtually every major UNIX implementation. While RISC chips like SPARC and IBM's Power are still hosting the operating system, Intel is working hard to ensure that its architecture will be the primary platform for UNIX computing.

But why should a company already so successful hosting Windows want to bother Windows' only viable competitor OS? Part of the answer is that Intel, for all its success, has not held on to its single-source position as well as Microsoft has. While there are no Windows lookalikes, Intel has had to contend with competing chip makers---most notably, Advanced Micro Devices---whose Pentium clones have sometimes out-performed the original. That has taken a toll especially on Intel's retail sales, where the company's market share has dropped to 54 percent, according to the research firm Computer Intelligence, versus 84 percent at the same time the previous year. These figures put increased pressure on Intel to succeed with its new 64-bit architecture, the IA-64, whose first chip, the Merced, is due in mid-2000.

In addition, both Microsoft and Intel are---like other technology companies---extremely opportunistic, willing to make business deals wherever they can find them. In a climate of shifting business alliances, the Wintel alliance is no exception. Thus Microsoft has ported Windows to the Alpha RISC chip while Intel is eyeing markets beyond the PC platform. "Intel has always been pragmatic," writes Dan Gillmore, the technology columnist for the San Jose Mercury News. "An Intel investment in Red Hat---and therefore in Linux, an open-source variant of the Unix operating system---isn't a loud declaration of independence from Microsoft. But it's at least partly a signal, and not the only one."

Gillmore sites other signs of independence as well. Intel has invested in Be Inc., whose high-performance operating system first began on the Macintosh, and in Freegate, whose server running FreeBSD gives small companies a fast path to the Internet. Indeed, the Internet itself---with its fierce tradition of open standards---is one place where Microsoft NT is not playing well. Intel sees an opportunity there, as well as in the appliance market, where the company is playing with its StrongARM chip design, which it acquired from Digital Equipment Corp.

But the main reason Intel is blessing UNIX is that it cannot afford to take chances with its IA-64 64-bit architecture, which will "raise the bar" on high-performance computing, especially on the server side. IA-64 represents Intel's future, and while clone makers try to reverse engineer it, Intel will for a while have the market to itself. Presumably, the complications that have put off Merced's scheduled 1999 shipment will also present barriers to the clonemakers---or at least Intel hopes so.

Meanwhile, Intel has made no secret of its product roadmap. The company plans a two track development process: one for the current 32-bit IA-32 product family (which ranges from the Intel386 to the Pentium II); and a second for the IA-64 product family of chips that maintain full IA-32 binary compatibility but are aimed at servers and workstations. After the Merced's ship date slipped a year forward from 1999, Intel felt it necessary to disclose its plans beyond that. The company said it would follow the Merced 18 months later with the "McKinley," which will run at double the speed. As of October, Intel said it has finished Merced's microarchitecture definition and plans to ship samples to OEMs in mid-1999. (At this writing, details of the Merced's design, including speed and cache size, remain undisclosed.)

If servers are the primary market for the Merced, then server operating systems are very important to Intel. And so while Intel, being no fool, has pledged allegiance to Windows NT 5.0, the company can't help but have noticed that the newest NT version is late and mired in the difficulty of trying to debug enormously complex code. Like the Merced chip itself, Windows NT 5.0 is a project of enormous breadth. Even NT 4.0 reportedly has more than 35 million lines of code, about three times that of Sun Solaris. At least until the NT monster is tamed, UNIX---an OS that some have declared dead---is a rock of stability that Intel wants to support. As an Intel spokesperson put it to me recently, "If there's a promising commercial operating system out there, we want it to run on our processors."

So with that as background, let's look at Intel's most visible areas of UNIX support.

Buying a piece of Red Hat

Perhaps the most surprising news regarding Intel and UNIX is the company's equity stake in Red Hat Software, which along with Caldera, is a leading Linux distributor. Red Hat's president, Robert Young, said that the importance of the deal was not so much the money as the new partnership. Investors like Intel and Netscape lend Red Hat the respectability it needs to compete in the all important corporate market and get the attention of IT consulting firms who recommend products to their corporate clients. "Red Hat previously was not perceived as someone they should spend cycles working with," Young said in a CNET interview. "With Intel's and Netscape's endorsement we expect that to change."

In announcing the investment, Intel said it would form a technical and engineering liaison to the Linux community, working with the I2O special interest group and Project UDI (Uniform Driver Interface---see below) "to enable greater availability of Linux device drivers for Intel-based servers." Intel also announced the formation of an Intel Linux user group. The Intel stake will also put Red Hat higher in the information chain, enabling it to get important design information about forthcoming Intel processors ahead of their release, a privilege enjoyed by large software developers like Microsoft, as well as developers of binary-only UNIX versions. "We need the relationship with Intel because the vast majority of our users are running on Intel processors," said Young. "This opens the door for closer cooperation, both for us and the whole Linux community, with Intel. This is why Intel is interested in investing. They want that ability to deploy their technology into the Linux marketplace on a much faster basis."

Of course, Red Hat can certainly make use of the revenues from its new investors and will plow it largely into technical marketing and support. "In order for us to be able to convince an MIS director to deploy Red Hat Linux systems on an official basis, we have to be able to deliver a level of enterprise support that, without denigrating our current support effort, we simply don't deliver at this point," Young said.


IBM's collaboration with SCO and Sequent has been widely seen as an attempt by IBM to catch up in the UNIX market. The deal calls for IBM to adopt SCO's UnixWare 7 as its 32-bit UNIX operating system for the Intel server market, with IBM contributing parts of its own AIX technology. The two companies plan to merge the two OS versions to provide a single UNIX application platform for Intel and IBM's Power RISC processor. For the IA-64 architecture, the two companies will collaborate on a 64-bit UNIX operating system based on AIX, UnixWare and Sequent's PTX OS, in order to compete with Solaris 7. To that end, IBM and Intel are pooling their resources to establish a multi-million dollar fund that will assist software companies to deliver middleware tools and application programs. ISVs pledging support included Informix, Netscape, and Novell.

Sun Solaris 7 on IA-64

The collaboration between Intel and Sun was announced back in December 1997 when the two companies said they would cooperate to optimize Solaris for the Merced processor. Sun said it would establish a "Solaris on Intel" porting and tuning center for ISVs and OEMs to tune and optimize applications. Sun Microsystems' president Ed Zander has boldly said that the company has a three year lead on Microsoft, a projection industry analysts think is optimistic. None the less, Sun appears to be gaining market share against NT, as well as other UNIX competitors, and is considered the revenue leader for commercial UNIX developers. (SCO is the market share volume leader.)

Pushing Project UDI

Intel has been playing an active role in Project UDI (Uniform Device Interface), which would create a common interface framework for UNIX. The resulting middleware would enable peripheral developers to write one device driver that could support multiple versions of UNIX. Companies active in Project UDI include hardware vendors Adaptec, Bit3, and Interphase, and UNIX vendors Compaq-Digital, Hewlett-Packard, IBM, Santa Cruz Operation, and Sun.

John Miner, who heads Intel's Enterprise Server Group, told an Intel developer forum that Project UDI would significantly improve the availability of devices supported by UNIX operating systems, provide for a more standard, robust device interface and device environment, and improve the performance of device drivers deployed in the UDI format. "Intel and the UNIX operating system vendors will be delivering to the industry a UDI Developers Guide, as well as interface specifications, and will be providing source code for this UDI environment to the free source, open source Linux community," Miner said.

ncurring Microsoft's wrath

It's difficult to imagine Microsoft applauding Intel's involvement---particularly with its Red Hat investment. While any grumblings have not come to light, leaked Microsoft memos indicate concern about the open source movement in general, and Linux in particular. "The Linux OS is the highest visibility product of the Open Source Software (OSS) process. Linux represents a best-of-breed UNIX, that is trusted in mission critical applications, and---due to its open source code---has a long term credibility which exceeds many other competitive OS's. Linux poses a significant near-term revenue threat to Windows NT Server in the commodity file, print and network services businesses."

In addition, the U.S. government's antitrust suit against Microsoft has uncovered friction between the two companies, particularly in Intel's aborted development of Native Signal Processing (NSP) software that would have initially supported only Windows 3.1. "Intel was wasting its money by writing low-quality software that created incompatibilities for users,'' said Bill Gates in videotaped testimony. Steven McGeady, vice president of Intel's Content Group, testified that Gates was particularly upset about the lack of Windows 95 support, and threatened not to support forthcoming Intel technologies, including MMX, Intel's multimedia technology. "The threat was both credible and fairly terrifying," McGeady said . "If we kept pissing them off ... they weren't going to support MMX." He said that Intel was reluctant to give Microsoft more advance notice of its NSP project because its engineers worried that Microsoft would try to "stomp it out of existence....It was the fear that was eventually realized."

McGeady also characterized Microsoft's Internet strategy as "embrace, extend and extinguish:" embrace an emerging standard like Java, add proprietary extensions, and hope that the original product gets extinguished. In its cross-examination, Microsoft tried to discredit McGeady as a disgruntled employee who could not get NSI to market and blamed Microsoft for his troubles-an allegation McGeady denies. Whatever the truth, it is clear that the working relationship between the two companies has not always been an extended honeymoon

An interview with Michael Pope, Intel's director of server software programs

Do you mostly see UNIX primarily residing at the server level?
Yes, definitely. The biggest case in point was at the Solaris 7 announcement, where Sun spent an hour and a half describing the features and benefits of the Solaris 7 product without one word about the workstation or desktop. That to me is an indicator that UNIX is a server operating system.
Why are you supporting UNIX on the server?
When you look at all server segments, we definitely see a need to support and have other operating systems besides Windows NT on Intel architecture in order to meet the entire range of server environments, from server appliances all the way on up to the highest end data center servers.
But you've done more than simply say we're glad that UNIX is there. This is an aggressive strategy.
First you need to understand what Intel's growing role has been in working with the software community. Go back 10 years and you'll find a lot of UNIX support efforts inside Intel at the engineering level. We've a long standing relationship in our involvement with SCO. We were among the early co-developers of the ABIs with USL [Unix Systems Laboratories], predating even Novell's purchase of it. So when you look back over the years, we've had a lot of UNIX input. That's especially true when we were coming out with new processors-the 386 and 486-and we wanted to make sure that there was a common ABI. If you go back about six or seven years ago, we had published a common ABI specification working with SCO and Novell.
Where did the your group come from?
We formed the Enterprise Server Group about four or five years ago, recognizing the importance of software in the server market. As our systems became more enterprise ready, we saw the need to expand our ISV program beyond just operating system vendors to database and application vendors, and doing more than just low-level software enabling. Over the last several years, we've established application solution centers-where in very short bursts of time we can do optimizations, say, between an operating system and a database on an Intel platform.
Did Intel initiate Project UDI?
It's somewhat of a derivative of SCO. They had an effort going, and we picked up on that. We worked on it, supported it, and worked with the other vendors to get them to understand the benefits of it.
Were you in a better position because you are not a UNIX vendor yourself?
Yes. In a lot of cases the vendors like working with Intel in that respect because we don't develop an OS itself. We're more of an enabler.
How does the agreement between IBM, SCO and Sequent to produce a single UNIX affect Intel?
Look at the background. The highest volume, most successful shrink-wrapped operating system out there today is still Novell Netware. Since then, Windows NT has come on very strong, becoming the preferred shrink-wrapped operating system for application servers. By preferred, I mean the volume leader.
We think there is room for a shrink-wrapped volume UNIX leader. If you take the combination of SCO's volume and shrink-wrap capabilities, their operating system runs on almost every single OEM server platform. If you combine that with IBM's enterprise expertise and commitment to port all of their middleware and their own internal applications, we think that combination can create a leading candidate for UNIX on Intel.
Did you play matchmaker?
We had some discussions with them, but ultimately it was driven by the business decisions both at SCO and at IBM.
Where do you see Linux?
Linux is very rapidly growing at the low end of the server market. Its capability is growing daily in the breadth of applications it can serve. We are now seeing a lot of third-party robust, enterprise-oriented application vendors such as Oracle and Netscape commit to putting their applications on top of Linux. So we see Linux as another high-volume operating system in the server space. As such, we are interested and eager to provide support for that community.
Why did you invest in Red Hat?
It's part of a broader Intel ISV program where we will do selected investments. We invest in software vendors that we think have a strong potential for delivering the added value into the user community on Intel architecture. Red Hat fits this model.
The president of Red Hat, Bob Young, said that it wasn't so much the investment as it was the clout of the partners, meaning Intel and Netscape.
Obviously there are benefits on both sides. One of the benefits that Red Hat gets is a proof point that their strategy is right, which they can use as a selling point. In general, smaller companies in these transactions can gain a "halo effect" through association with a successful company like Intel or Netscape.
I think it's fair to say that Microsoft has been jealous of its business relationships. Do you walk a fine line here between your support for UNIX and your relationship with Microsoft?
We have a great relationship with Microsoft. We're engaged in a lot of areas, and the fact that we do something like invest in Red Hat doesn't stop us from working with Microsoft, as well. We're very bullish that Windows NT is going to be very successful, and is going to sell a lot of servers on Intel architecture. We also think, however, that the server market is expanding very rapidly and there are other interesting areas that Intel architecture can and should play in, that require us also to dedicate and devote some resources and investments.