Jim Dennis: Open LetterIn response to: http://www.internetwk.com/columns/logic061499.htm
JP MORGENTHAL
June 14, 1999
Look Before You Leap Into Linux Adoption
It seems I cannot pick up a paper these days without seeing Linux touted as the greatest thing since sliced bread. Upon reading these rave reviews, I can't help but think that most of this admiration is emerging from a revulsion to Microsoft Windows, as well as the relatively low cost for the Linux operating system compared with other versions of Unix.
Personally I think Linux is "the greatest thing since" UNIX. It has the features of UNIX with a refreshing idealism and energy --- to finally provide interfaces that are attractive to general audiences and users (look at KDE and/or GNOME --- those are just the beginning, first cuts of things to come).
Some revulsion to MS Windows is to be expected. You can do it the MS way (with their look and feel) or you can't do it. Many people revolt when they are given little or no MEANINGFUL choice (i.e. control over the appearance and operation of their systems).
The effect of low cost cannot be discounted. MS-DOS 5.0 could be had for about $50 retail. Win '9x is almost twice that. NT is about six times that (for the stripped down version). Linux is free (you can get a copy from a friend to save you the long download times) --- and a couple of million copies a year are being sold for prices ranging from ~$2.00 (http://www.cheapbytes.com, among others) to $200 (Caldera's deluxe package?).
The only way that the Microsoft line of operating systems got where they are today is by being the cheapest (specifically by hiding their costs into the hardware bundle --- in a way that was clearly anti-competitive and probably in violation of U.S. trust laws). The fact that early PCs were simply not powerful enough to run the versions of UNIX that were contemporary to them made UNIX irrelevant for the first 10 years after the PC was introduced. (Irrelevant to to the PC and broader microcomputer markets, that is).
The Macintosh largely suffered from it's "luxury" pricing; and still does. No platform has undercut PC pricing, and no OS has been offered on an equal economic footing to those provided by MS.
But I think it's critical that we, as a community, keep a perspective on the impact of Linux for the following reasons.
I agree that all user communities should keep perspective in all things. I don't know which community you are referring to, here.
Linux is an open-source project; therefore, all changes to the kernel are subject to review and approval by a small team that controls this portion of the operating system. Companies that add features they need, but that are not accepted into the core distribution, may find themselves in a redevelopment and retesting cycle every time a new version of Linux is released.
Let's compare this to the situation of that same company if it relies on a closed source operating system. Here our hypothetical company has no recourse since they CANNOT "add features they need."
They can't add the features they need unless they convince a software company (a small "team" that controls ALL access to the OS). If our hypothetical company does convince this software company to add custom features they have NO realistic guarantee that said software publisher will continue to support those patches in future versions. (Remember MIPS and Alpha versions of NT --- where are the users that relied upon Microsoft's support for these platforms).
So, in the Linux world our hypothetical company can make the patches they need. They need no permission and need not negotiate with anyone to gain access to the sources. They can release their patches into the world or keep them private. (If they want to keep the patches private they are constrained by the GPL from distributing a derived product --- that is selling the resulting kernel. However, they can make internal/private use of their proprietary patches without restriction).
If they release their patches the developer community (in this case the Linux kernel team) might accept or reject them. Regardless of whether their patches are accepted they are free to distribute their patches and to release products that are built therefrom. Other organizations and developer teams are then free to accept these patches and those groups and the original hypothetical company then benefit from by building their own (probably smaller) community or faction to maintain and upgrades these patches.
You are correct. Some companies might find themselves in a development and testing life cycle as new versions of Linux appear. They need not go through this process "every time a new version is released." Well written kernel patches will generally be easy to maintain and test. The interfaces among major kernel components don't change quickly under Linux. With the introduction of the 2.2 version of the kernel these internal interfaces are likely to be even more stable in future releases.
So the this criticism boils down to:
"If you have any choice you may find that others don't agree with the choices you make."
... and you're apparent preference for NT in this regard seems to imply that:
"we should therefore not accept any choice in these matters."
(I'll leave the theological ramifications that one might gain by expansion of this proposition to your imagination).
Windows supporters still outnumber Linux supporters because Microsoft provides a better value proposition.
You seem to be paraphrasing one of Microsoft's chief propagandists with this phrase "better value proposition."
It's a buzz phrase.
In terms of sheer user base the predominance of MS Windows is a historical artifact. MS-DOS predominated on the low-cost microcomputer hardware; MS has phased MS-DOS out through their strategic upgrade, release and pricing policies (and a bit of strongarm manipulation of key channel purchases during the mid to late eighties).
Since most vendors still ship MS operating systems as the default OS --- and often the ONLY choice on 90% of the world's most affordable computers (generic PC clones) --- they still have the dominant marketshare.
The fact that Linux has gained up to 17% of the server market and possibly as high as 5 to 7 percent of the overall PC market is simply astonishing. Currently Linux seems to be doubling it's installed base every nine months or so. That rate will probably flatten out in a few years. Of course that could very well mean Linux on 80% of new systems three years from now.
(No, that's not a prediction on my part. That's a simple extrapolation from the estimates over the last few years. Of course: "Historical performance is no guarantee of future results.")
I'm not sure how you arrive at figures to compare MS Windows "supporters" versus Linux "supporters." Not every user of either OS is a "supporter" by any means. If you measure user satisfaction you'll probably find that MS Windows has among the lowest.
This is not actually an attack on MS Windows itself. I'm not saying that there is any inherent reason for people to prefer Linux or MacOS over MS Windows. However, more people are "forced" to use Windows than any other OS. The minorities that use Linux, MacOS, FreeBSD, or other operating systems generally do so by their own choice.
Inevitably many of the people who use MS Windows do so because "it came with the system" and/or because their management "require it as the corporate standard."
So polls on user satisfaction would not reveal anything about "inherent software quality" or popularity.
Based purely on user satisfaction (ignoring pricing factors), we'd probably see MacOS, NeXTStep or BeOS in the top spot.
(I personally don't like MacOS, so no personal preferential bias compells me to make this observation).
... Windows NT Server Enterprise Edition ships with a full complement of Internet services, including Web, proxy, index, messaging, database, transaction and firewall services. With Linux, these services will soon be available as a multivendor product. Whereas Microsoft's products are designed to work with one another and the operating system's services, users may spend a significant amount of time trying to integrate these components under Linux. The most critical of these integrations will be security and access control.
Every general purpose Linux distribution ships with a 'full complement of Internet services including Web (Apache), proxy (, SOCKS, DeleGate, etc), index (glimpse), messaging (sendmail, qmail, etc), database (mSQL, mySQL, PostgreSQL), and firewall services (ipfwadm/ipchains, IPSec and others)' or makes them readily and freely available.
(I left "transaction services" from this list because I'm not sure what NT's transaction service APIs are like, and therefore can't name the analogous Linux/UNIX counterparts. There may not be any, or there may be some that are commercial, or others that are free and that I simply have not yet discovered).
Linux is just beginning to be retrofitted for symmetric multiprocessing. Without robust SMP, Linux servers can support only small companies and single applications. If you're managing multiple servers for increased scalability, you're better off using multiple NT servers all participating within the same domain.
You are wrong. Linux is not "just beginning to be retrofitted" for SMP. You are also wrong regarding this perceived SMP imperative. Purchasing agents at large and mid-sized companies are constantly told that they "need" SMP --- frequently they don't; but they lock themselves into applications architectures that do require SMP (often without understanding the technical issues involved).
Linux version 2.2 is SMP. It is not a "retrofit" but is the result of years of extensive work. The development cycle between Linux 2.0 and 2.2 was the longest in the history of the platform. The adjustments in kernel locking granularity were a major portion of that work.
Linux SMP is scalable to at least 4 processors. We'll probably see kernel scaleability to 8 processors by the end of this year.
These issues of scaleability are far more subtle then members of the press are willing to discuss. SMP/kernel scaleability is highly dependent on the applications involved --- and on the ratio of userspace computation to system call events in particular. Thus computationally intensive applications with relatively little I/O and relatively infrequent need to request memory allocatioins (the system call) will still benefit from the SMP support in Linux 2.2 (or even in version 2.0 with it's single global kernel lock) even as we go beyond 8 processors.
In reality scaling beyond 8 to 16 processors is probably a waste of time and resources. At that point we are probably far better off persuing clustering technologies (with varying degrees of coupling over various high speed, low latency communications channel such as "Firewire" fibre channel or even gigabit/s ethernet.
The fact that a few Linux "Beowulf" clusters appear on the world's "top 500" list of supercomputers (http://www.top500.org) is old news. The form of loosely coupled clustering used in these systems is in no way Linux specific --- and similar clusters could be built over any form of UNIX.
However, Linux and the FreeBSD family of operating systems provide researchers and academics with an unprecedented opportunity to advance the state of the art in clustering --- and companies (and governments) will be free to incorporate these advances as they need them.
Also, remember that Linux is still Unix. One of the reasons for Windows' growth has been the complexity of configuring and maintaining Unix operating systems. Simply because a low-cost version of Unix is now available, it does not automatically generate more people capable of managing and configuring these systems.
This implies that you believe that a "less complex and more easily configured UNIX would have retarded the growth of MS Windows." I realize that this will seem to put words in your mouth but the reasonable conclusion is that you believe that UNIX is a superior operating system in some ways (save for its complexity and apparent management challenges).
I won't disagree with that point. However I'll have to take issue with your next statement.
The existence of a low-cost, widespread and popular form of UNIX does, indeed, automatically generate "more people capable of managing and configuring these systems." In fact it appears that the existence of Linux as approximately trebled the total number of people who can manage and configure UNIX-like systems.
This is not to imply that every Linux user is a competant system administrator. That is far from the case. In fact, among the many versions of UNIX, Linux probably has the lowest percentage of professionally skilled users. Many experienced UNIX users prefer FreeBSD or one of its brethren. Some Linux users consider the adoption of FreeBSD to be a "graduation" to a more traditional version of UNIX.
(Of course many professionals with extensive cross-platform experience, such as me, prefer Linux --- while still others are operating system "agnostics").
However, the fact that there are at least a few million Linux users, virtually all of whom have installed the OS by their own choice, suggests that there are many people who are aquiring the base skills to be UNIX professionals.
If only 1% of the Linux users approach professional competancy then we still gain tens of thousands of new sysadmins and consultants and integrators available to the world marketplace every year.
Your comments here also simply ignore (or display your ignorance of) the administrative user interfaces that are now available for Linux (like the linuxconf utility that's included with Red Hat 6.0 and available for other distributions of Linux, and the COAS suite that's included with Caldera, and portable to many forms of Linux and UNIX).
All UNIX has ever needed to make it more accessible has been interface programs. Writing user interfaces for a stable, robust platform is not challenging. Every commercial UNIX has included GUI and menu driven interfaces for common administrative functions for several years. IBM's SMIT (for AIX) is considered to be exemplary among many UNIX professionals.
The Linux community may surpass the efforts of IBM, HP and other corporations by sheer demand. There are already more Linux users than there are installations of all other forms of UNIX combined. There is now great incentive to make Linux more accessible and easier for the masses to use. (Traditional UNIX markets are often atagonistic towards GUIs --- a feeling which I share with the most stalwart curmudgeons among them. However, Linux is not controlled by such people).
Personally I don't like linuxconf. I haven't played with COAS or Webmin (a web browser, forms driven, administrative tools for Linux). However, I expect that these packages will be more than adequate to serve the needs of most novice administrators --- and the experts will still be free to bypass them and simply slice and dice the configuration text we often prefer.
Linux is a college student's project gone astray. The version that will be supported by Sun Microsystems and IBM on its hardware will fall far short of each of these company's own Unix operating systems in features and capabilities. If you're responsible for operating system selection in your company, be wary of the Linux play. Hey, I'm all for a competitor to Windows, just give me more than what Microsoft has to offer-not less.
This is remarkable. In what way has Linux "gone astray." If you'd said "run amok" I might understand. "Gone astray" suggests that there is some sheppard looking for his or her lost Linux. However, Linux is far from sheepish. It may be a bull in the china shop of commercial software.
The versions (note the plural) of Linux that will be supported by Sun, IBM, SGI and others will probably serve the needs of thost customers who select it. Those customers will have other operating system choices for the foreseeable future.
(It is worth noting that Linux users ported the OS to SPARC and UltraSPARC, Alpha, PowerPC (IBM RS/6000), and MIPS (SGI) platforms with little or no support from the principle manufacturers of these systems. The support of IBM, Sun, SGI, and others is welcomed by the Linux community --- but certainly not required. It is clearly in their best interests to broaden their market and Linux is an easy and incredibly economical way for them to do so; while gaining plenty of positive PR along the way).
Any features that Linux lacks may be added by anyone with the need and inclination. The whole point of Linux is that it encourages those who need more of it to co-operate and provide. That is the whole reason that RMS (Richard M. Stallman) created the FSF (Free Software Foundation) and the GPL (General Public License).
Sun, IBM, SGI and others have wasted enough of their resources in their long-running game of one-upmanship. Perhaps they see in Linux an opportunity to declare truce in the OS wars. Perhaps they can focus on the true "value proposition" they can offer to their customers (offering superior hardware with high degrees of interoperability).
I recommend that anyone who is responsible for the selection of operating system and platform for any company, organization or project be wary of industry pundits and the press.
Such decisions should be made based on a thorough requirements analysis. Anyone who picks NT because JP Morgenthal prefers it, or avoids Linux because JP Morgenthal says it provides an inferior "value proposition" (whatever that means) is just as much a fool as if they choose Linux because Jim Dennis likes it. They only judge of "value" should be the actual users of a system --- the customers for whom it is intended.
I can't help but wonder specifically what JP Morgenthal wants "more of." Does he want more cost for the software (perhaps he makes commissions or owns Microsoft stock)? Does he want more instability (perhaps he prefers for his consulting customers to be dependent on his services)? Does he want Microsoft to continue to push other software companies out of of business? (Maybe he's short-selling Corel stock, or whatever). Perhaps JP hasn't tired of the endless variety of MS Word Macro viruses, trojans, and worms to which Microsoft's systems perrennially fall prey.
Perhaps the specter of proposed strengthened legal force to those rediculous software licenses that prohibit fair evaluation and reporting (limiting the scope and content of software reviews) and the creation of compatible and interoperable products (restricting the right to reverse engineer protocols and file formats) is something that JP Morgenthal looks forward to. (I've referring, of course to the UCC 2B legislative proposals).
Personally I'm sure that I want nothing to do with any competitor to MS Windows that offers any more of these "value propositions."
Naturally JP Morgenthal could respond with a long list of features that MS Windows has or promises. He could then make the assertion that Linux and other forms of UNIX don't provide these features. I've seen such lists. However, lists of features inevitably assume that the market has homogenous needs. Many "features" are of benefit to only small segments of a market. Mass advertising and publication (the foundation of the entire computer industry press) is focused on features because it is impossible to truly reflect the real needs of individual users and organizations.
Heed your own counsel. Ignore the FUD and judge all "features" purely by their application to your needs. If it is not of benefit to you --- it is irrelevant at best.
JP Morgenthal is president of NC.Focus, which provides strategic planning, analysis and consulting of application integration technologies. He can be reached at jp@ncfocus.com
Jim Dennis is best known as the "Linux Gazette Answer Guy" He's a senior support analyst and instructor at Linuxcare (http://www.linuxcare.com). The opinions expressed here are purely his own and, in no way, reflect those of his employers (though they might be pretty close). Jim can be reached at jimd@starshine.org.
Contact Starshine Technical Services