Pacific Connection(英語)

Sun and the X86: an Uneasy Alliance

この記事を読むのに必要な時間:およそ 13 分

The notice on Sun Microsystems' website is so brief you could almost miss it. "The Solaris 9 OE [Operating Environment] Customer Early Access software is...only available for the SPARC platform. We are deferring the productization and release of the IA-32 platform version of Solaris 9 OE at this time."

Solaris 9 is the newest release of Sun's UNIX-based operating system, of which the Early Access Program is its beta testing program. Sun says Solaris 9 offers improved system management, Web application development components, and increased security through encryption and password authentication. Solaris 9 also includes "hundreds of Open Source commands, libraries and utilities," thereby supporting Linux applications. And, Sun says that the newest release will take advantage of hardware improvements in new SPARC technology, including the 900 Mhz UltraSPARC III.

Will none of these advantages be available on X86 platforms? In using the term "postponed," Sun has not entirely shut the door on Intel users. But with no announced schedule as to when an Intel port of Solaris 9 will be available, many observers suspect a change of strategy in how Sun addresses the low end server market. But given the market pressures, the real surprise, at least in retrospect, is that Solaris supported the Intel platform as long as it did. Sun first provided Intel support for Solaris in the early 1990s, around the same time it opened a separate division, SunSoft, to try and profit from its software offerings. But two years ago, with tough competition from Linux and NT, Sun made Solaris free (except for the cost of the CD-ROM) as an incentive for application development. That move was good for developers, but did not make a healthy business model, and it's hard to blame Sun for not continuing with Solaris 9.0.

Still, the sudden move has made a number of users unhappy. They have complained that Sun pulled the plug without any warning to users, that Solaris has been an important, low-cost alternative to Windows, and that, if the price were right, they'd be more than willing to pay licensing fees.

If Sun does end support, it will sever one more strand of its direct relationship with x86 platforms. Java, of course, runs on x86 servers, as does StarOffice, Sun's office application. While Sun itself has long stopped shipping x86-based servers, Sun Cobalt, the business unit Sun acquired in September 2000, does use the platform for its Linux-based "appliances" although the company has been careful not to mention it. Press material for Sun Colbalt's RaQ XTR server appliance, for example, names a "733 or 933 MHz CPU." More than ever, Solaris and SPARC are an inseparable combination.

Solaris: squeezed on three sides

The reason for Sun's move has to do with three technologies: Intel CPUs, Linux, and Windows NT. All three affect Sun's ability to differentiate itself in the server market---which, despite the prominence of Java, account for most of the company's revenues.

On the CPU side, the Sun UltraSPARC RISC processor is up against Intel and its formidable research and development team. Solaris 8 currently supports a wide number of implementations of Intel's IA-32 architecture, including the Intel Pentium, Pentium II, Pentium III, and Pentium 4, as well as the AMD-K series and AMD Duron. Noticeably missing from this list is support for the IA-64 architecture, and here, Sun has given mixed messages. Back in October 1999, it announced that Solaris 7 was running on "engineering prototype systems based on Intel's Itanium processor" and in February 2000, Sun reaffirmed that it would support Intel's current IA-32 platform while "extending support for the IA-64 platform.

"Sun and Intel have OEM customers who have made a significant investment in Solaris as their operating environment for the IA-64 architecture," Anil Gadre, Sun's Solaris Software vice president and general manager, said back then in a written statement. "Sun is fully committed to working with Intel to insure the success of our joint customers. The market is speaking clearly and strongly for Solaris and Intel. Demand for our Free Solaris on Intel Program is high and our Solaris OEMs have reaffirmed to us their commitment to offer the Solaris Operating Environment on their IA-64-based systems."

But in January 2001, Sun chastised Hewlett-Packard's move to drop its PA RISC architecture for the IA-64 platform, saying that the "strategy only ended up costing HP UNIX market share, which it is now desperate to regain....Sun, on the other hand, has maintained a consistent strategy that is based on an integrated hardware/software stack and binary compatibility for applications throughout its product line. Now the undisputed leader in UNIX servers, Sun has out-shipped HP 4-to-1 at the high end over the past four quarters (according to IDC). Customers continue to flock to Sun because they trust the SPARC/Solaris environment will be consistent." In other words, Solaris and SPARC are what distinguish Sun from Hewlett-Packard in the server environment.

As for Intel, the company is not surprised. "We have been seeing, and continue to see, the most momentum on the Intel- based servers from Microsoft, Linux and other flavors of Unix, with very little demand for Solaris on Intel architectures," said Bill Kircos, a spokesperson for the company, in an eWeek report. "As such, we are not surprised or disappointed by the move. I think one of the reasons for the lack of interest in Solaris on Intel was that it did not perform as well and may not have been as well tuned for the Intel architecture as it was for Sun's proprietary SPARC platform," he said.

Sun and Linux

By contrast, Sun has augmented its Linux support. Back in late 1998, Sun had announced a Linux UltraSPARC port. Last February, Sun announced it would ship a full implementation of Linux, that it would ship Linux on two distinct product families---the Sun Cobalt Appliances, and a new class of low-end single and multiprocessing x86-based servers, available mid-year, and will include both single- and multiprocessing architectures. Sun says its Cobalt line, whose low end units are priced at less than $1,000, have an installed based of more than a thousand. Sun will continue to add to that line beyond its two current configurations: the Qube, with an eight square inch form factor, and a 1.75 inch rack mountable configuration.

The company's current strategy with regards to the two operating systems is to bridge the gap by making Solaris compatible with Linux applications, while contributing to Linux development so that IBM is not the only big player involved. Solaris 8 already supports some Linux applications, and with Solaris 9, Sun says it will support additional interfaces, commands and utilities. The LinCAT (Linux Compatibility Assurance Toolkit) will simplify the process of assuring that Linux applications will run on Sun's Sun Fire family of servers. Sun has made some of its technologies available to the Linux platform, including the iPlanet directory, Forte for Java development tools, the Java/XML platform, Project JXTA, the Star Office suite of applications, Sun Chili!Soft ASP, and the Sun Grid Engine. Sun is providing tools to help insure compatibility between Solaris and Linux. Sun has settled on Gnome as the single graphical user interface for Solaris.

As for Windows 2000, Microsoft remains Sun's evil nemesis---and that's reflected in everything from new product development and marketing, to statements from company executives, and in legal battles both past and future. Sun may not appreciate IBM's success in positioning itself as the big league Linux company, but when it comes to Microsoft, the Middle East saying applies: "The enemy of my enemy is my friend."

Unhappy users

While not providing an x86 Solaris 9 port may make good business sense, it has angered users, some of whom have organized a website: An open letter to Sun CEO Scott McNealy circulated the site and called the announcement a "broken promise and a betrayal" and asked for reconsideration. As of this writing, it had nearly 800 signatures, including people from an impressive list of organizations, including universities, research labs, and corporations, including a few engineers who claimed to be from Sun itself.

The letter suggested that while the free download "succeeded in putting Solaris in the hands of over a million people," customers might well be willing to purchase the next release if it were priced comparable to other commercial operating systems. "As far as we can tell, no effort was taken by Sun to gauge whether this solution would be acceptable to the user community and we are concerned that a potential revenue stream may be ignored."

The letter argues that Solaris on the x86 platform complements, rather than competes with 64-bit SPARC solutions, providing a transition for installations seeking higher performance. "It is many of our experiences that Solaris x86 is installed on existing systems running the competitions' operating systems and closed applications. Once we transition these systems to Solaris and open solutions, Sun quickly proves to have the superior technology. Our customers are subsequently much more receptive to Sun hardware solutions as they outgrow their x86 systems. As developers and system administrators we value the ability to run the same operating system across the enterprise: laptop, desktop, workstation, workgroup server, and large-scale server. We revel in our ability to demo our enterprise applications to potential customers on off-the-shelf notebook computers. Finally, we believe that the widespread adoption of Solaris on the x86 platform has and would continue to increase the creation of hardware drivers and applications for both x86 and SPARC systems."

Some users have made their displeasure known on various websites, like "Like many others I'm both shocked by Sun's decision, and yet I understand the economics behind this move," said one poster, arguing that what's needed is better driver support, documentation and installation. "If they would simply build the product as described above, shrink wrap it, and push it through the local book store for $100 a copy they would have a best seller on their hands. Let's face it, a full function business class OS for $100 is a deal."

"Basically Linux is killing them," wrote another. "It won't be long soon before the Intel chips will overtake the SPARC chips in speed and reliability then that will sign the death warrant for Sun. I think that Sun should keep the Intel support and concentrate on enabling more Linux apps to run more easily on Solaris. That way they don't need a huge application base."

One writer said that vendors who "link the sale of their operating system to the purchase of their own proprietary, over-priced hardware have a difficult time maintaining market dominance... Microsoft employs many tools to keep the market penetration that they enjoy, and one of the major reasons for their success has been the illusion that Windows runs on an industry standard hardware platform....All of that said, I don't foresee any large scale abandonment of Solaris due to this announcement, nor do I see a rush to switch to Windows on those few Intel based Solaris machines. The only OS that stands to gain ground in this one is Linux."

Sun: Solaris is not going away

The obvious conclusion is that Sun has now permanently ceded the x86 market to Linux. If you want Solaris, by a SPARC machine. And why not? Linux has hundreds of volunteer developers, is making steady progress in supporting higher-end server functions, and is already seen as a viable alternative to Windows 2000, at least in some applications. Sun will thus regroup to save business resources, supporting its proprietary version of Unix on its mostly proprietary chip---and therefore differentiating itself from the pack.

Nobody at Sun is quite denying that. But what about the next step: dropping Solaris and supporting Linux throughout the product line. "People have asked, 'Are you shifting from Solaris to Linux?' I can't stress enough the answer is no," said Sun's Stephen DeWitt, vice president and general manager of content delivery and edge computing. A joint statement by DeWitt and Gadre said that Sun will support Linux in two areas: edge computing (data caching) and embedded systems. Presumably, the mainstream server business-where Sun still gets most of its revenues---will be driven by SPARC and Solaris. At least for the time being.

Sidebar: A Conversation with Sun's Graham Lovell and Herb Hinstorff

In online discussions over Solaris on the x86, Sun's point man has been Graham Lovell, director of Solaris product marketing. He's the one who made the announcement on behalf of the company and is mentioned prominently in's open letter to Scott McNealy. I spoke with Lovell, and his colleague, Herb Hinstorff, who manages Sun's Solaris/Linux compatibility efforts.

Where is Sun in terms of Solaris on the x86?

Graham Lovell:
We have a strong product in the marketplace right now, Solaris 8, that we've been shipping for two years with an Intel version, which has had seven revisions. With Solaris 9, there are no more planned updates to Solaris 8. But under our normal life cycle model, we would continue to sell and support Solaris 8 for quite a while. We see at least another two years of life left in Solaris 8 on Intel platform, so we'll continue selling the product for that period. Then we will be supplying support for two to five years after that.

Is Solaris 8 still a giveaway on the x86?

It's still freely available. There's a cost for the media kit, but you can register the licenses for free on the website. If you want to five copies, you buy a copy of the media and register your five copies on the web, and you get five copies for the cost of one media kit.

Will Solaris support on x86 end with Solaris 8?

The message we've been passing on for the last few months is that we are deferring the productization of Solaris 9 for Intel. We don't feel now is the right time to incur the costs of productization for the product.

Some users suggest you should go back to a fee model on it. Is that something you've considered?

We've got a user community out there that is so excited about Solaris on Intel that they are prepared tell suppliers that "instead of giving this away for the cost of the media-why don't you charge us between $100 and $200 each time." As a marketer, that's music to our ears. However, there are costs associated with doing this. We are continuing to ship the product that we have in Solaris 8. We're getting high volumes of that product going out to the community. We're continuing to discuss with the community the longevity of Solaris on Intel.

Does it make sense then for X86 users who are interested in Sun to look to Linux as the future?

Herb Hinstorff:
People [first] look at the operating environment they want to deploy, [then] consider what platforms to deploy it on. When people look at Solaris, they see its scalability, manageability, availability, security, all built in, plus the fact that it's a truly global product. Those are they reasons they're attracted to Solaris---regardless of platform.

But if you're not continuing to support it on the x86....?

We haven't made any decision to support it or not with Solaris 9. We've just deferred the productization of it. We're continuing to make Solaris 8 available and continuing to support that product. And we have the option to productize the Solaris 9 and ship it to customers if and when we feel that's the right time.

And certainly there's a strong, vocal Solaris-on-Intel architecture community that wants to see us do more.

In Cobalt, you already have a Linux appliance. In mid-year, you're announcing a general-purpose Linux server. What's the difference?

An appliance tends to be a dedicated function box in a system that you specifically buy and get supported to do one task, whether it's caching or web service, or firewall protection. The general purpose server is one that is set up and supported for developing and running generic applications.

Another example. If you look at something like the TiVo box [which records television shows on disk] our understanding is that that runs Linux. But the user is totally unaware of what the operating system is inside TiVo, nor do they care. Whereas when you are on a general purpose box, you are exposed to the operating system in all of its glory, so you actually do really care. You care about what applications can run on that and how well it runs on the box.

So that's the difference between what Cobalt is doing and the forthcoming server?

Yes. Cobalt has been focused on single-function devices, and with this announcement, we're saying that we will expose more of the general capabilities of a Linux server.

You have worked to make Solaris 8 and Solaris 9 more Linux compatible. How much progress have you made?

Solaris 8 is already extremely compatible. Both Linux and Solaris share a common heritage. As our CTO likes to say, "they both came from the same swamp," so they employ very similar methodologies. They both use the POSIX API set. They both run basically the same set of open source applications. The similarities are far greater than the differences. It's good for Sun and good for Solaris to have that happen because Linux, because of the momentum it's enjoyed especially on the low end, has really provided a new wave of momentum and innovation around UNIX.

We've produced a disk worth of common libraries, which provide the application programming interfaces, as well as common utilities and tools, including the entire GNU development environment with Solaris. In addition, we've made the Java platform available on Linux. We support the Java 2 Standard Edition as a tier one platform on Linux as well as Solaris, so that it comes out at the same time on Solaris, Windows and Linux, providing another common platform for applications to be developed.

Some observers believe that as the line between SPARC and Intel blurs, Solaris will be the differentiation point for Sun. You'll buy Sun in order to get Solaris. Do you agree?

People don't buy chips, they don't buy OSs, they buy systems. What Sun sells is an integrated set of systems, storage and service to provide a solution to our customers. Clearly, Solaris is a differentiator for Sun, one of the attributes of the system that has been very successful in the market and has a huge set of tested, qualified commercial applications. At the same time, there are many attributes of the SPARC architecture and our systems architecture that provide the overall value to customers.

Is Microsoft the common enemy?

I think Linux and Solaris are natural allies in the battle for open standards, open interfaces and an open methodology as opposed to Microsoft, which tends to promote their own products and facilitate locking around their own products. Solaris is based on open standards, open interfaces, and always has been. Solaris and Sun have been very much in front of the market in terms of promoting open interfaces, community-based development. We have a very long track record, dating way back to the early '80s and our earliest roots with BSD.

Anything you want to add?

One other thing is the work Sun has been doing over the last 2.5-3 years to support the Linux and Open Source community, not only in packaging products for Linux like the Forte for Java, which is the leading Java development environment for Linux, and StarOffice, which is the leading office suite for Linux. We've also contributed a huge amount of code. We're the largest corporate provider of intellectual property to the open source community. We have made the entire development base for StarOffice, for Forte for Java, for the JXTA peer-to-peer protocol, for the grid engine product---all available as open source. We've engaged the community on development. We're living the collaborative community lifestyle and making a big advance to those open source efforts and to Linux as a result. We are also very large contributors to other organizations, such as, and Gnome.

If that's all true, wouldn't it be a fair assumption over the long term that Linux will be the operating system of choice for Sun?

You'll find that Linux is an operating system that is going to end up on a lot of Intel-based hardware, because of its availability and that fact that it actually does run and is very popular, especially for people who may be coming out of school or doing research or prototyping. It's there. You're going to find it. What Solaris provides is a very rich environment for deploying applications, and a huge amount of commonality with Linux. Gartner [the research firm] says that 98 percent of what you get in a typical Linux distribution is not the Linux kernel or Linux drivers, but the generic open source tools and utilities, which are both common to Linux and Solaris.

So it's the ancillary code that counts the most?

Right. A lot of our value add for Solaris is in the technology and in the know-how inside the kernel to do things like scalability. Sun is the sharing some of that expertise with the Linux community.

You're saying that Linux is behind in a multiprocessing environment?

Not behind, but less mature. "Behind" has an emotional connotation which I don't think is right for them. Linux's focus has been on commodity hardware, low numbers of process accounts, etc. We've been focused on higher-end systems, on commercial applications. They are not behind. They're ahead in a number of areas where commodity-based platforms and small process accounts are concerned.

Moving forward, the real platform for development going to a higher level: middleware, the Java-based infrastructure and Sun One. That's going to be common across both Linux and Solaris.

Are you saying that the real conflict here is between Microsoft's .Net and SunOne?

Yes, it is absolutely that. The Holy Grail for developers is that they can run on any system, anywhere. They can run on the 250 million PCs a year that are sold, on all the UNIX-based systems, on mainframes---providing the largest possible marketplace for your software application. SunOne and .Net are both trying to capture mindshare from developers. The difference is that SunOne can deploy across almost any architecture at all, including Microsoft Windows, whereas their .Net is restricted just to a Windows environment.


Bart Eisenberg

Bart Eisenberg's articles on the trends and technologies of the American computer industry have appeared in Gijutsu-Hyoron publications since the late 1980s. He has covered and consulted for both startups and the major corporations that make up the Silicon Valley. A native of Los Angeles and a self-confessed gadget freak, he lives with his wife Susan in Marin County, north of San Francisco. When not there, he can sometimes be found hiking with a GPS in the Sierra, traveling in India, driving his Toyota subcompact down the California coast, or on the streets of New York and Tokyo.


1980年代後半より,『Software Design』や『Web Site Expert』などの雑誌に,アメリカのコンピュータ業界のトレンドと技術に関するレポートを執筆しています。シリコンバレーで,スタートアップ企業から大企業まで幅広い分野でコンサルタントを務めました。

ロサンゼルス生まれで,自称ガジェットフリークです.現在,妻のSusanとともに,サンフランシスコ北部のMarin County在住。また,SierraのGPSを携えてハイキングしたり,インドを旅したり,カリフォルニア海岸をドライブしたり,NYや東京の街中を歩いたりしています。