Linux in the Enterprise: Fact vs. FUD

When I first started touting Linux as a soon-to-be superior alternative to Microsoft Windows, almost no one at my company had even heard of the product. Nearly two years later, it's difficult to find a computer magazine that does not extoll the virtues of Linux. However, these praises are often laced with caveats: Linux is a "server OS," it's difficult for novices, it's "not ready for the desktop." To some extent these concerns are simply due to a reasonable fear of the unknown. However, lately we are beginning to see more FUD ¹ being deliberately spread about Linux, very little of which is based in truth. Here is a link to a superb example, the Microsoft "Linux Myths" page:

http://www.microsoft.com/NTServer/nts/news/msnw/LinuxMyths.asp
(Note:This has been removed from the Microsoft website.
Here is the referenced article: Locally archived copy)

The purpose of this essay is to address some of these concerns, to demonstrate why Linux is a viable alternative for Windows 98/NT as both a server and a workstation, to demonstrate that Linux is here for the long-haul, and explain why it will inevitably become a dominant force in the operating systems market.

The FUD

Microsoft's Linux Myths page gives several "reasons" why Linux is a poor choice for the enterprise, all of which are based on inaccurate or out-of-date information. Some of them are simply wishful thinking on Microsoft's part.

1. Linux performs poorly compared to Windows NT.
2. Linux's stability is only based on anecdotes.
3. Free doesn't mean low TCO
4. Linux security is weak.
5. Linux makes no sense on the desktop.

Let's look at them individually:

Linux performs poorly compared to Windows NT:

It is a fact that some benchmarks have shown that Windows NT performs some tasks better than Linux². It is equally true that other benchmarks have shown the reverse³. Any operating system can be tuned to perform a task better than any other⁴.

However, what isn't stated is often telling. For instance, the tests that Microsoft points to are ostensibly to measure "Common Customer Workloads." Yet they measure the performance of a 4-processor Windows NT machine against both a single-processor and 4-processor Linux box. How many "common customers" use 4-way NT boxes? Very few, in my experience. NT loses these same benchmarks when comparing single-to-single processor and dual-to-dual processor machines. While it is true that some customers require heavy hardware to get the job done, with 16, 32, or more processors, such systems are in the minority, even in the shops where they are used. Where they are used, they are more likely to use a high-end Unix than NT to do the job⁵.

Microsoft also claims that NT performs better than Linux when serving static web pages. However, e-business is not powered by static pages. It's powered by Active Server Pages and CGI⁶. Windows doesn't fare nearly so well in these comparisons. Also, Microsoft, when charting throughput of Internet Information Server vs. Linux+Apache carefully refrains from mentioning that it would take at least 5 incoming T1 lines attached to your Linux server before this scalability becomes a factor. How many "common customers" have 5 dedicated T1 lines feeding into a 4-processor server? Such a setup is not what I'd call "common."

As for the technical specs quoted by Microsoft... they are out of date. The Linux kernel addresses 4 GB RAM, not two. The largest file size depends on the hardware and file system you choose⁷. Linux supports many file systems⁸; NT supports far fewer). And the number of filesystems is growing. Recently SGI announced that they would release their XFS file systems to Open Source, and would be embracing Linux. XFS, when released, will allow a maximum file size of nearly one million terabytes⁹

The 2.2 Linux kernel removed the 128 MB limit on swap file size, you can mount as many as eight. The maximum useful size of a swap area now depends on the architecture. It is roughly 2GB on i386, PPC, m68k, ARM, 1GB on sparc, 512MB on mips, 128GB on alpha and 3TB on sparc64. However, most users do not use swap files at all; they use swap partitions.

SMP has been supported since kernel release 2.0, and has been incrementally improved since then; even the most recent enhancements are publicly available.

Note that not all of the features supported by Linux are included by default in every distribution, but they all can be added if missing. This is by design, and by choice. A distribution that is designed for a home workstation or laptop may leave out certain features that only make sense on a high-powered server. Symmetrical Multiprocessor support, for example. The fact that this can be compiled into the OS kernel or not, depending on the needs of the users allows every installation to tailor the smallest, fastest, most stable custom kernels to their specific needs. It also demonstrates one of the most important ways in which Linux is scalable and Microsoft Windows is not.

Speaking of scalability, Linux is capable of distributed clustering ("Beowulf" technology¹⁰), which can enhance system performance dramatically. Several of the world's 500 fastest supercomputers are in actuality Linux Beowulf clusters¹¹. You'll require special software to take advantage of Beowulf clustering, but if you need a supercomputer it's guaranteed that it won't be running Windows.

Understand that while Linux is not the correct choice for every server application (in my opinion, no operating system fits that criterion) it's becoming increasingly hard to find an application for which it's not the best fit.

Linux's stability is only based on anecdotes.

According to Microsoft, Microsoft's anectodes are "testimonials." Linux testimonials are "anecdotes." Be that as it may, there are a lot of these anecdotes. Many of them are include documented uptimes ranging from months to years¹².

Few people argue the fact that Windows NT's stability is poor. It is common practice among NT administrators to periodically reboot their machines to prevent unscheduled downtime. Note that this practice doesn't prevent the downtime at all, it simply re-classifies it as "scheduled!" In a list of enhancements to be included in Windows 2000, Microsoft includes system reliability¹³. If they admit the need to fix it, they're admitting it's broken.

Microsoft also claims that Linux has no journaling file system. They're almost right. XFS is still being cleaned up for the Linux port. Ext3fs is still in alpha phase. However, ReiserFS¹⁴ is in late beta, and has a journaling extension.

Microsoft claims that there is no "commercially proven" clustering technology for Linux. In fact, TurboLinux does includes failover clustering.

Also, when reading FUD, it's helpful to spot the "weasel words." Here, "commercially proven" is the phrase to note. Because Linux is new to the marketplace, these words embody fear, uncertainty, and doubt in a two-word package.

Free doesn't mean low TCO

Actually, it does. Microsoft's TCO calculations are based against other commercially marketed Unixes, which have very expensive initial acquisition and support contract costs and traditionally high education costs¹⁵. Typically you can expect to upgrade Microsoft products every two to four years (depending on Microsoft's variable release schedule), at considerable cost not only due to re-licensing, but also due to training and implementation, and in fact the upgrade costs of Windows products is quite high¹⁶.

On the other hand, Linux has no acquisition cost. It has no upgrade licensing costs. Once initial learning curve has been met, the learning curve for upgrades is remarkably flat. Support costs can be very low, as purchased support can be supplemented with award-winning Usenet support. Improved stability and uptime minimize the costs associated with downtime. With the exception of the kernel, any component of a Linux machine can be upgraded without even re-booting or downing the machine. In situations where a machine is doing multiple duties (mail and database server, for example), the mail server can be upgraded while the database server continues to support your business.

Microsoft complains that Linux companies will charge for support. In fact, it is reasonable for them to do so. Microsoft itself charges for support for a product that they licensed you at considerable cost. However, with Linux you're not required to purchase support at all. While your support options for Windows are limited, your support options for Linux are not. The Linux User community, operating at no charge, garnered the 1996 InfoWorld Product of the Year award for Best Technical Support. It should be painfully obvious that the editors of InfoWorld passed over Microsoft and all other commercial vendors when presenting the award. Remember that you must consider the quality and the efficacy of the support as well as the cost. Although you can purchase local support for Microsoft products, such support is strictly limited to training and workarounds. Linux support is now available from such firms as Red Hat, LinuxCare, and VA Systems, among others. Because Linux is Open Source, support firms are capable of providing you with permanent solutions, including kernel patches and enhancements. Microsoft Windows support is simply not in the same league.

Add together the facts that acquisition and upgrades are cost-free; that award-winning support is free; and that purchased support of superior quality; and that the improved stability of the OS itself requires less attention, it's obvious that Linux is cheaper to own in both the short and the long run.¹⁷

Linux security model is weak.

Microsoft's entire argument regarding Linux security is inaccurate. They claim that Linux provides access controls only for files and directories. In fact, devices and processeson a Linux machine are presented to the kernel as files¹⁸ and are subject to access control.. Linux can also support ACLs¹⁹.

Microsoft claims that security is all-or-nothing, that administrators cannot delegate administrator privileges without making an ordinary user a full administrator. This is blatantly wrong: individuals can be assigned to groups having only the required privileges²⁰.

Ed Curry, the man who led Microsoft's efforts to obtain their C2 security rating has publicly stated his opposition to the rating, as shipped versions of the software are not the same as those that were tested.²¹ Microsoft's Knowledge Base confirms that this is the case.²²

Microsoft claims that your security administrator must be an expert to properly configure security. My own knee-jerk reaction to this is, "When do you NOT want an expert supporting your systems? If you can't afford one full-time you hire a consultant." NT's security is easy fairly to set up, but it's not as easy as pressing some "make me secure" button somewhere. Properly configuring Linux security is simply different. It's largely a matter of removing those services that are not needed and religiously applying security patches as they appear. This should be standard operating procedure for any shop, regardless of the platforms used²³.

Linux makes no sense on the desktop.

This isn't what Microsoft told the Department of Justice²⁴. In court and under oath Microsoft officials maintain that Linux is a threat to Windows dominance. Today the claims are different.

Microsoft claims that Linux does not support USB, Plug and Play, and Power Management. In fact, it supports all three, although you need to download USB support separately²⁵.

Microsoft claims that existing Linux GUIs are cumbersome and difficult to use. In fact, my mother sat down and began using the KDE²⁶ desktop with no training, no prior experience, and not one single problem.

The enterprise situation is very different from home use for a number of reasons, not the least of which is that corporations often want to lock down the desktops. Ask a system administrator what his major challenges are and you'll find that high on the list is preventing users from haphazardly installing unauthorized applications on corporate desktops. Corporations need to control the desktops, the applications, the hardware purchased, and the network holding it all together. They also hire qualified system administrators to do the job; and to meet their goals they require a system that is easy to use and to maintain. "Easy" in this case has a new definition: It does not mean that anybody can configure it like an expert. Instead, it means, "somebody else set this up for me and I never have to touch it again because it never breaks." This is the "business definition" of easy, and it contributes to Linux's low TCO. My mother is more fortunate than most in that she gets the benefit of this "business definition" at home. I know many people who wish they had the same for Windows.

As far as the availability of applications²⁷ goes, beyond the functions provided by general purpose programs like StarOffice, WordPerfect, and Netscape Communicator, most business applications are proprietary; many are written in-house. They also operate in a controlled environment. Corporations typically order uniform hardware configurations. And while it's certain that Linux does not support every possible combination of hardware, it's equally certain that in most businesses, this simply doesn't matter. It's not difficult to avoid hardware incompatibilities (and such incompatibilities are not unknown to Windows. It's one reason why corporations order uniform configurations.)

For business use, the major general purpose tool Linux lacks at the moment is a native Lotus Notes client²⁸. Notes mail is readily accessible through the IMAP or POP mail clients provided by StarOffice or Netscape Communicator, or via Web mail through a browser. Note that in the absence of a native client that Notes databases can be web-enabled, and the Windows Notes client can be run if you install WINE (a set of libraries providing Windows compatibility).

I don't think that Linux is right for every business, nor for every desktop. But if 20% of the business users in your company require Windows, It makes precious little sense to pay for the remaining 80% to have it. The more users you have, the less sense it makes.

The Future

So far we've discussed the current state of Linux, and we've noted that it's a robust, stable OS that can meet the needs of many businesses today. But what about the future? Will it suddenly disappear? What if Microsoft gets into the act... can they "embrace and extend" it to death, as they've done with other technologies? In short, is this just a fad or a dead end?

The truth of the matter is this: Linux won't go away because it can't. The legalities of the GNU General Public License²⁹ make it impossible for any one person to monopolize the operating system, and there is no legal mechanism to un-GPL a program. What is the GPL? Basically, it's the copyright license that makes Linux free. The GPL states that the source code for Linux must be made available free of cost, and that any changes made to the system must be shared under the terms of the GPL. This discourages the kind of fragmentation that has occurred with Unix in the past. As a result, although there are dozens of different Linux distributions³⁰, any of them can run a binary that's been compiled to run on any other, assuming that you have the necessary libraries (similar to DLLs) installed.

Sometimes people who misunderstand the nature of the Linux community attempt to spread FUD by inciting in-fighting among the proponents as follows:

> Don't you guys think that Linux might in the end just cannibalize Unix
> and not kill M$? There have been a number of articles and studies
> hinting on that. The most recent being the Gartner Group article.³¹

First of all, this is based on the mistaken assumption that Linux developers want to "kill M$." They don't... what they do want is to build a better operating system: one that is robust, stable, and user-friendly. The Linux philosophy is that given enough people working in the same direction, you can have it all.

Now for the meat of the argument. A recent Gartner Group report quoted by Microsoft³² indicates that Linux's main impact will be to "speed erosion of weaker Unix variants." I assume that the Gartner Group was being fair-handed when they made the statement, and in fact I agree with them in the short term. However, as Joe Barr recently observed, the Gartner Group knows nothing about Linux³³, as they've had very little incentive to learn about it. Their studies are not funded by Linux concerns because that's not where the money is³⁴. And these studies are based on polls of commercial concerns that do not yet have a clear understanding of Linux themselves.

What Microsoft attempts to do by quoting the the statement is to capitalize on fears that the de-fragmentation of Unix is a Bad Thing³⁵. This couldn't be further from the truth.

Take IBM for example. Why are they throwing their weight behind Linux³⁶ if it competes with their own AIX? It's because IBM doesn't sell AIX, per se... it sells servers and provides AIX to make those servers work³⁷. it doesn't make a lot of economic sense for IBM to develop and sell an operating system if there's a functionally similar OS that's free for the taking, and which is developed and supported by a mass of developers vastly larger than their own. Now they can engage in the far more cost effective practice of simply optimizing Linux for their own hardware and can market that hardware on its own merits. The same is true to some extent of other vendors, like Sun. By embracing Linux, Sun can concentrate on the high end server market having freed resources from developing an OS for their low-end servers. SGI, the developer of Irix, has released their journaling file system to Open Source³⁸, and are sponsoring the retail release of the Debian Linux distribution³⁹. They would not do such a thing if it were not in their best interest⁴⁰.

Secondly, Linux is more likely to strengthen Unix. Why? Because people like what they know. As Linux penetrates the home and academic markets a new generation of programmers will emerge who know and like Linux. They will find that they can move from Linux to Unix with negligible effort. In the past it's been tough to find Unix programmers because learning Unix has been expensive while learning PCs has been cheap. That's changed. PCs used to ship with one programming language... BASIC. Specifically, Microsoft BASIC⁴¹, and on the strength of this market penetration Visual BASIC and VBA are of strategic importance to Microsoft. Now PCs ship with no language at all (unless MS Office shipped with your PC, in which case VBA sort of counts). You have to buy a language on your own, which takes desire and money... lots of it. Red Hat Linux, on the other hand, ships with over a dozen programming languages, all of which are free. On which system are you more likely to learn to program? To companies like Sun, Linux provides a cheap low-end training arena for Solaris developers.

Thirdly, even if other Unixes were cannibalized, what would it matter? Linux would remain, and the point I made in the above paragraph works in reverse. Those Unix developers that now exist will move to Linux with no effort, and there will be no discernible effect in the workplace. Unix would be unified and there would be no fragmentation to which FUD-mongers could point. Linux would still continue to penetrate the home market, creating developers as in my previous point, and Linux would resume eating into Windows market share after the slight pause to belch.

The second point is so important that it bears repeating: Microsoft Windows is an enterprise contender only because it penetrated the home market first. People used it at home and as a result of their experience they wanted it at work... and when it first appeared in businesses, it was not the better solution. A similar process is currently occurring with regards to Linux. Programmers and System Administrators are quietly putting it in place because it works, and works well, with little to no intervention on their part. A new generation of programmers is using it because they're students and Linux fits anyone's budget. These people will want Linux, and they will have it. Users are asking for it because it some of them have become convinced that Windows has a DLL specifically designed to measure your proximity to deadline and increase the likelihood of a crash accordingly. They want something stable and easy... and they want the business definition of easy.