Open Source--The Unauthorized White Papers Chapter 5, Linux vs. Windows

Open Source: The Unauthorized White Papers

Chapter 5

Linux vs. Windows

Newspapers and computer magazines love to describe Linux as the Windows killer, but as we have seen in Chapter 5, Linux is not yet ready to cross over into the mass market and find a place on the average desktop. There is little point in comparing it with Windows 3.x (16-bit) or with Windows 95/98 (32-bit). Microsoft is pushing users to the new Windows 2000, is hoping to be able to field a 64-bit Windows for the coming Itanium chip, and is trying to enter the embedded market with Windows CE. Linux will be stiff competition in both the 64-bit and embedded platforms (see Part IV, "Linux is Moving UP… and DOWN.")

For those businesses that are not vendors of 64-bit or embedded systems, the only significant Microsoft versus Linux discussion is about servers (or workstations); that is, Windows NT/2000. Among servers, the Linux displacement threat is probably greater to low-end UNIX rather than to NT, but even if Linux is not displacing NT servers, it is often purchased in place of an incremental NT server. This chapter will look at inroads Linux is making into Microsoft’s server market, and the countermeasures Microsoft is taking in return.

The Competition with NT

Business users never anticipated a contest between Open Source and NT because for all of the early history of Linux, they were unaware of its existence, even as it was being smuggled into their backroom operations. Rather than wait for an influential visionary to spot the emerging technology, the innovating technologists themselves introduced the pieces they needed, bringing their home hobby into the workplace. Because expenses were low or nil, and because the Open Source software did its work quietly and efficiently, IT managed either to remain ignorant, or preferred not to ask questions. From Web server to file and print server to glue among heterogeneous systems, Open Source software offered inexpensive customizable solutions to the technically proficient.

Web servers

Linux first entered most companies as the platform on which to run the Open Source Apache Web server; we can track its growing use by looking at the Netcraft Web Server Survey (http://www.netcraft.com/survey/). The survey publishes the relative shares of the various servers used on the World Wide Web. Admittedly this survey is not scientific, as it depends on gathering sites not by random selection, but by taking URLs entered by visitors, and then regularly polling this growing list. The survey is thus skewed by self-selection, and by the fact that one server may host multiple polled Web sites, so that a single server may be counted scores of times. For this reason it does not do much good to worry about fine distinctions; the large picture is interesting enough, however.

Apache (presumably running on Linux or BSD) is on approximately 62 percent of servers surveyed, while Microsoft’s IIS server (running on Windows NT) is on about 21 percent. There is also a survey for third-party certified SSL server sites: these commercial sites return the result that about one-third are running Microsoft and something fewer than one-quarter are running Apache. But if we assume that the Stronghold C2 SSL servers are running on Linux, Linux’s share rises to slightly more than Microsoft’s. In any case, there is serious competition between Linux and NT in this space.

Servers in general

It is always difficult to gather information about the spreading use of Linux. One copy will do for many machines, the copy need not even be paid for, and many Linux users will pay for each new version of the system as it appears. Researchers can count installed systems, and the operating systems installed on them, and they can count new systems as they are sold with the operating system already installed. They can also count box sales of Linux distributions.

Sales of New Servers

International Data Corporation (IDC; see http://www.idcresearch.com) studies of unit sales in network operating systems (NOS) provide the figures listed in Table 6-1.

Table 6-1

Market Shares for Network Operating Systems

Operating System

1997

1998

1999

Linux

6.8

15.8

25

Windows NT

36.0

38.3

38

NetWare

26.4

22.8

19

UNIX

20.7

18.8

15

OS/2

6.3

3.0

 

Other NOS

3.8

1.3

3

Above Totals

100%

100%

100%

From 1997 to 1998, Linux’s share of the server market OS sales increased from 6.8 percent to 15.8 percent, more than doubling its share. If we accept a market-size estimate that sets growth at some 26 percent (from 3.5 million to 4.4 million), that means sales of some 695,000 copies of Linux distributions. NT Server weighs in at 38.3 percent (nearly equals 1,685,000) of the 1998 market. In that year Linux and NT were the only two operating systems to increase their market shares, while UNIX and NetWare were both having their shares cut into by the newer operating systems. By 1999 NT market share was stalled, while Linux nearly doubled its market share and NetWare and UNIX continued to lose ground.

The figures cover only U.S. sales of Linux by the major distributors (Red Hat, SuSE, Caldera, and so on), and they do not cover the free downloading of these distributions over the Web, or the purchase of them in inexpensive two and three dollar copies from Linux Mall (http://www.linuxmall.com) or Cheap Bytes (http://www.cheapbytes.com/). Nor do we know how many copies of the distributions were not used for servers, but simply for standalone machines. About all we can tell from the figures is that Linux is a popular new technology, and that NT is hardly in danger of disappearance.

Server Installations

The Gartner Group’s DataPro Information Services (http://www.gartnerweb.com/public/static/datapro/main.html) published a "User Ratings Survey of Operating Systems" in November 1998, which indicates the installed base of servers among respondents. Most of the respondents were in consulting and professional services, and all were in the United States. Of these, some 67.7 percent had Windows NT installed, some 12.5 percent were using Linux, and 3.9 percent were using BSD. (Of incidental interest are the findings that, on the desktop, some 89 percent had Windows 95 installed, but only 17 percent had Windows 98 installed. The survey also reports that users are buying Windows NT like hotcakes, while not having a high opinion of it as an operating system.)

Respondents indicated their plans for increased or decreased usage over the next three years; the results show that Linux is the UNIX-type operating system most named for increased use among general (that is, including Microsoft) users. Growth in NT is fueled mainly by replacing lower-end Windows systems, by replacing NetWare, and to a lesser extent by replacing UNIX. The DataPro survey also indicates the uses for Linux. These are summarized in Table 6-2.

Table 6-2

Uses of Linux in All-Linux Businesses

Use

Share

Web server

33.3

Applications server

9.9

Scientific and technical workstation

14.8

Networked workstation

9.9

Desktops

6.2

Enterprise

9.9

Total

84%

DataPro explains the 9.9 percent figure for Enterprise use by stating that Linux-only operations tend to be technically oriented smaller enterprises that appreciate the savings that Linux enables.

As for betting on Linux among resellers, the Computer Reseller News poll published in October 1999 reported a rising confidence that "Linux would be a viable alternative to Windows in the small and mid-size company markets within the next 12 months," and that the increase in respondents expressing similar confidence in Linux for the enterprise market was even more dramatic. (See Table 6-3.)

Table 6-3

Computer Reseller News Poll of VARs

Company Markets

August 1999

September 1999

 

Yes

No

Undecided

Yes

No

Undecided

Viable for small and midsize in next 12 months

45

   

52%

37%

11%

Viable for Enterprise in next 12 months

26%

45%

29%

45%

28%

27%

A growing number of VARs are making up their minds (only 11 percent were undecided about Linux for small and midsize businesses), and there was a surprising jump from August to September 1999 in the number of VARs who believed that Linux would be viable for their enterprise in the next 12 months.

Values — Linux versus NT

There are several Web sites that compare Windows NT and UNIX (including Linux and BSD) in detail, such as the one put up by John Kirch (http://www.unix-vs-nt.org/kirch/). We have already discussed relative costs in Chapter 4, and it is only a question of whether we go into technical matters in greater or in lesser detail. Because both systems are improving, any discussion will quickly become dated, but the larger points of difference include the following:

* Mainly because of stability, cost comparisons are favorable to Linux. Additionally, a typical Linux distribution has everything in one box; for NT, additional products are needed.

* Linux, without a GUI or with a lightweight GUI, will run comfortably on low-end hardware that NT cannot run on.

* Linux can emulate an NT server by using free server software called Samba, enabling integration of Linux and NT without licensing fees, expensive hardware, or instability problems.

* The BSD OSs are older and currently more able to handle high transaction loads than either NT or Linux, and may be used to run Linux software and integrate with NT systems.

* Fewer servers are needed to accomplish the same tasks than with NT.

* Linux systems are easier to configure and support than NT.

* Linux runs on many platforms, including 64-bit Alpha from Compaq; NT is now confined to Intel.

* Open Source software may be modified by the user; unneeded portions may be dropped.

On the whole, users of large systems understand that neither Linux nor NT is a neck-and-neck competitor with the older and more functional UNIX systems; the question is which of the two newcomers will close the gap with UNIX more quickly. Microsoft recently announced its determination to bring Windows 2000 up to the mainframe level, and Linux is able to use distributed processing to emulate a supercomputer, but currently neither could safely run a large enterprise such as Sabre, the airlines reservation system.

Microsoft Slaps at Linux

Nineteen ninety-nine was the year of the Linux breakout, and the celebrity and successes of the upstart operating system led Microsoft to tell the Department of Justice that the rise of Linux and office applications for it was proof that Microsoft had genuine competition. In the marketplace, however, Microsoft was finally moved to notice the newcomer and to scoff at it.

Benchmarking

In October 1999 Microsoft struck out at Linux by posting NT/Linux comparisons on a "Linux Myths" page on its Web site (http://www.microsoft.com/ntserver/nts/news/msnw/linuxmyths.asp). A number of the alleged shortcomings are related to the usual benchmark problems — Has the problem been set to favor one server over another? Did the tested systems receive the same opportunities for tune-up? — but some of the charges have been heard before, have merit, and the Linux community is on its way to answering them.

The outrage at the Linux Myths page was not nearly so great as the initial outburst at the publication of the April 1999 Mindcraft, Inc. tests of Linux versus. NT sponsored by Microsoft. Linus Torvalds initially denounced the tests, but then decided that the better approach was to "Admit it: we sucked," and show the superiority of Open Source development by tackling the shortcomings head-on. At USENIX in June and at LinuxWorld Expo in August he admitted that too long an interval had been allowed to pass between stable kernel releases, and that kernel development was "developer-driven," rather than user-driven. He then allowed that "the anti-Linux group at Microsoft is just a user in disguise," and that such "bug reports are good: they just don’t go through normal channels." Finally, Torvalds promised that kernel releases would be more frequent in the future.

Note

Linux releases are odd-numbered for "developer," that is, experimental versions, and even-numbered for stable releases. The gap between 2.0 and 2.2 was two-and-a-half years.

In addition, the summer of 1999 brought a number of Open Source gifts to Linux in the form of donated code to support needed improvements. These contributions demonstrate the increasing interest by large companies in Open Source; and the successful integration of these firms into the Open Source world will demonstrate the viability of the Open Source development model. Among the shortcomings complained of in Linux are the lack of a journalizing file system and the inability to address memory larger than 2GB. A month after the Mindcraft tests, SGI, Inc. announced that it planned to make available the source code for XFS, a 64-bit file system that would also work on 32-bit systems. This will enable a future Linux kernel to access file systems up to 18 million terabytes, and file sizes up to 9 million terabytes.

Just before the Linux Myths page hit the Web, Siemens Information and Communication Products, and the Linux distribution SuSE announced that they had developed a memory extension allowing Linux to access up to 4GB of memory on Intel servers (Linux could already do this with Alpha servers), and that Torvalds was already planning to include the technology in a future kernel release. Open Source development serves both the community, by making its software stronger, and the hardware manufacturers who donated the code, by growing the Open Source market and enabling the manufacturers to sell their products upward into the more lucrative markets of database-intensive uses.

As Linux moves along to expand its functionality, especially in the fields most often criticized, it will be interesting to see whether the current projects to provide Symmetric Multiprocessing (SMP) and high-availability clustering will also benefit from code donations from large companies.

The Halloween Papers

The Halloween Papers serve as the Pumpkin Papers of the GNU Generation, purloined proofs of a dark plot to destroy the democratic sunshine of Open Source and to bring it and everyone under the power of Microsoft. They are in fact two confidential Microsoft memos. There has been heavy discussion of them on the Web, and while there has been less said about them in the press, they hang like background radiation in any discussion of the Open Source future.

The first memo was slipped to Eric S. Raymond a couple of months after it was written. Raymond says that he rushed to write a commentary on it, working into the late hours of Halloween 1998 and publishing the next day on the Web (http://www.opensource.org/halloween/); by adding a commentary he hoped to avoid possible legal action by Microsoft. The second memo, a follow-up to the first, reached Raymond a few days after the publication of the first.

Under the heading "Halloween Documents," Raymond has posted heavily-annotated versions of the two Microsoft memos, links to press coverage, and Microsoft’s official responses to the leakage and content, countered with rebuttals in the acid and playful style characteristic of Open Source controversy. The whole collection is worth studying for what it tells us about the perspectives and intentions of both Microsoft and the Open Source community, and how they work. From the Microsoft documents we learn of their high regard for Open Source products, their comparison of Linux to Java as threats, and what weapons they will use to combat this latest enemy.

A Look Into Microsoft’s Strategy

The memos’ internal praise for Linux and its performance is both surprising and gratifying; there is even a wistful admiration for the Open Source process that produced them, along with the realization that Microsoft will be unable to experiment with it, even internally. There is a fear that the process is so attractive that, particularly when powered by the Internet, it will capture "developer mindshare" and divert interest from Microsoft products. Linux, which within a year Microsoft will publicly deride on its Linux Myths page as based on a 30-year old technology, is here covertly admired as the beneficiary of 25-plus years of UNIX experience, and Open Source applications are recognized as having "commercial quality." While the memo believes that Linux is a more immediate threat to UNIX, it also perceives a near-term threat to NT-server revenues because hardware manufacturers will be able to put out inexpensive special-purpose servers using Open Source software (as indeed Whistle and Cobalt were already doing).

The memos could not go on to discuss the strategy of combating Open Source until they defined the nature of the threat. It was immediately identified as comparable to that of the Java VM (Virtual Machine), but harder to combat because the enemy was a process, and not a company. Furthermore, because Open Source code was easily available from multiple sources, Open Source applications could not die easily, and thus enjoyed "long-term credibility." The alarming conclusion was that Open Source could not be defeated by FUD. FUD (the sowing of Fear, Uncertainty, and Doubt about competitors’ products or their viability) is old stuff in the world of computer competition, and Microsoft has not done anything that IBM did not do in the days when it was mighty in the land. But Open Source developers, as lovers of elegant technologies, loathe the intrusion of sordid sales and marketing practices into what should be purely technical decisions.

The memos did see that Open Source could provide a market-entry advantage by supplying cheap giveaway products that could swamp a competitor, and predicted that cooperative efforts (such as Red Hat Advanced Development Labs’ contributions of Open Source code to the community) would last only as long as vendors felt too weak to compete with each other. Once firmly established, they would introduce proprietary twists to their products. Furthermore, a vendor could use an Open Source project to preempt or forestall competitors’ projects by absorbing community talents for its own Open Source project. Thus, the Microsoft analysts managed to see the Open Source process as a continuation of the proprietary wars by other means.

Note

In the Halloween Papers, Microsoft noticed that "IBM is taking a lead role in optimizing Apache for NT"; among the "Worst case scenarios" covered is "IBM Adopts Linux?" (see Chapter 13, "Companies and Projects").

Further high-level analysis led to Microsoft’s assessment of what Open Source developers regard as one of their greatest strengths as, instead, a fatal "core weakness" suitable for direct "attack." This is the use of open standards for protocols that make the Internet possible (HTML, TCP/IP, and many more). The tactics to deal with them had already been tested successfully . . . on Java.

As we will see in Chapter 8, "The Drive Toward the Mainframe," Sun is still smarting from having included Microsoft as both a Java licensee and as appointed developers of the official reference implementation of Java on Windows — "Who could do it better?" they must have asked themselves. The whole point of writing Java code was that it could be interpreted on any platform that had a Java Virtual Machine (JVM) running; although the use of interpreted rather than compiled code made the software run slowly, Java’s ideal was "Write Once, Run Many Places." Microsoft decided to "embrace and extend" the Java standard (on Windows, at least) by making available to it all the Visual Basic or ActiveX libraries that developers had already written and that were already running successfully.

This extension was offered, understandably, as a benefit to programmers. The problem was that the resulting Java code would no longer run on a non-Windows JVM; if it ran at all, it could not possibly include all the benefits gained from calling the ActiveX components. Microsoft’s vehicle for this hijacking of Java was its J++ development tool, which made calls to the Microsoft Foundation Classes rather than to the Java Foundation Classes. Developers had no warnings about which of the functions they were using were tied to the Microsoft platform, and thus could not count on writing portable Java code when using the tool. J++ has since disappeared, having done its part to drive Java applets from Microsoft clients. Today Java runs very successfully on servers, although few are aware of the fact.

The Halloween Papers talk dismissively of open protocols as commodities. The tactic here is to "change the rules of the game" by inventing new protocols and by extending the functionality of other protocols so that Open Source projects either cannot use them or at least cannot enjoy their extended functionality; the optimal case is to "deny entry" to the Open Source projects. Specific Open Source projects are targeted, and file systems will eventually be included (Linux can read Microsoft File Allocation Tables or FATs, for instance) as the Microsoft operating systems take over middleware functions. The operating system will serve as a barrier to accessing resources.

The other tactic discussed — potential suits over intellectual property — "remains to be investigated." This ominous phrase reeks of FUD. The best use of the threat to sue is not the threat itself; it is the refusal to say whether you will or you won’t. Having studied the writings of Eric Raymond (which state that Open Source works best on projects that have some running code that functions and can be tested, even if buggy), the memo writers conclude that Open Source developers are not capable of real innovation, and will always be chasing to catch up with superior operating systems (including NT). Further, Open Source developers will be unable to ape their superiors without falling into violations of patent or copyright. No further details are given.

Although Microsoft denies that the Halloween Papers are anything other than a middle-management exercise, Eric Raymond has identified two of their endorsers as members of the highest management team at Microsoft. The Halloween Papers lead the Open Source community to believe that they will have an eventful walk down the road ahead.

How to Close an Open Standard

Microsoft included in Windows 2000 the Kerberos security technology developed at MIT under the X License. This license makes the technology free for all comers to use and modify as they please (http://web.mit.edu/kerberos/www/). Microsoft added its own material to a data field left blank for developers to use, and did not publish the changes. Although Microsoft acted within its rights, its refusal to document the changes serves to close an open standard and hampers smooth communication between UNIX and Windows 2000 systems via Kerberos security.

The other shoe dropped when Microsoft, responding to public outcry, announced it would publish the standard on the Web in May 2000. Those who responded found themselves downloading a self-extracting compressed file that opened with a non-disclosure agreement (NDA) requiring consent before revealing the information. The NDA warned that Microsoft was imparting a trade secret that might not be published or shared except under NDA. The required secrecy effectively prevented the implementation of the Microsoft standard in open standards or code.

It did not take long for the hotheads at Slashdot to post not only links to Web sites containing the secret, but the secret itself. Microsoft then sent legal notice to Slashdot demanding the removal of the secret, and of links to the secret. Slashdot refused, and appears to be ready for a long march through the courts.

The trade secret claim is not likely to prevail, since once a trade secret is out on the Web, it is a secret no longer. Microsoft should have taken better care of it than to post it on the Web; a sly developer can easily access the secret directly without passing through the NDA. According to Sam Byassee of Smith, Helms, Mullis & Moore, the copyright claim might not even prevail if a judge rules that the matter is factual, descriptive, and so uncreative as not to merit copyright. If the material turns out to have valid copyright protection, however, then even fair use is no defense for publishing the material, secret or not, on the Web.

Slashdot is invoking the First Amendment, saying that to remove postings on its site would violate free speech rights by forcing them to censor posts by their readers. While copyright would certainly overrule the free speech claim for posting the (former) secret, the case is less clear that merely pointing to another site constitutes a copyright violation. Additionally, Microsoft is demanding removal of readers’ posted suggestions on how to access the document while ignoring the NDA (a process probably obvious to most developers). The fact that Microsoft has invoked the new Digital Millennium Copyright Act (DMCA; see Chapter 14, "Intellectual Property") increases the complication because the Act has not yet been tested. Microsoft has at last found a way to go to court with a community it finds a growing threat.

Next chapter: The GNU GPL and the Open Source Definition

Table of Contents

Copyright © 2000 by IDG Books Worldwide, Inc. Published under IDG Books Worldwide, Inc. Open Content License.