Softpanorama (slightly skeptical) Open Source Software Educational Society

May the source be with you, but remember the KISS principle ;-)

GNU not Unix: Unixification of Windows as an alternative to Linux

Dr. Nikolai Bezroukov

The phrase "GNU not Unix" have a very special meaning indeed, probably unanticipated by Richard Stallman  ;-). For me often (but not always) using GNU utilities and other software ported from Unix to Windows is easier and provide more consistent environment then usage of native Windows tools. Unixification of Windows means maximization of the usage of command line tools and scripting.

It is not necessarily always means usage of  ported classic Unix utilities. Many UNIX utilities are outdated to the extent being obsolete (find, grep, sed: in most case using Perl is easier then messing with all obscure options that those utilities got for the last 30 years, unless you are professional Unix admin) and other (including native Windows) command line alternatives are probably preferable. But many are still extremely relevant (awk, expect, Korn shell, bash). Also for some Unix command line applications a better native alternatives exist in Windows.  For example, FAR is definitely better then port of Midnight Commander.

One of the tremendous advantages of Windows over Unix is that Microsoft managed to enforce a single macro language (to be more exact a pair of macro languages: VBA and JavaScript)  for many of its tools (including major workhorses like FrontPage, Ms Word, Excel, Visio,  etc). It is also used in some third party tools like Lotus Notes) and as an internal scripting language. In this respect only OS/2 can compete with Windows. At one point Sun has chances to match Solaris with Windows by using TCL but they never managed to pull it off. 

Windows has a powerful capabilities on the command line that are severely underutilized by most users. Actually Microsoft supports developers well and provides a lot of free or semi-free utilities. Windows Resource Kits are one such example. They are essentially collections of pretty powerful command line utilities. Unfortunately they are not very popular and are definitely underutilized.  I see creating your own powerful collection of command line utilities as an optimal way of using Windows in both personal and enterprise environments and Resource kit utilities as SFU are important components of this collection.

A powerful command line environment which is so characteristic for Unix can be more or less successfully replicated in Windows and provides the same advantages: a better, more "scriptable"  environment especially to use in tandem with Unix/Linux server (my preferred configuration).

I am against using exclusively Linux as some open source enthusiasts suggest. For businesses with Windows installed base the cost of migrating to a single new platform is huge, and there must be tangible and quantifiable benefits that I do not see.  Several recent papers and reports present conflicting views on the real cost of moving to Linux. Some of these reports (often dismissed by Linux enthusiasts merely because they were sponsored by Microsoft), point out that there are substantial costs to absorb for organizations that implement a pure Linux-based infrastructure.

Rob Enderle, a TechNewsWorld columnist, has aptly suggested that the mixture of beliefs, ideals and enthusiasm surrounding Linux makes "Linux only" crowd very similar to a young religion.  IT decision-makers should consider facts rather than beliefs. Let us remain objective about things and not be carried away by enthusiasm. And, by all means, let us avoid the disturbing fanaticism and religious fervor of those who are motivated by hatred of Microsoft.

While at the university environment Linux/Unix can (and should) be used as widely as possible, if might be not such a good idea to put it on the desktop.  There are still a very strong reasons to use Windows as a desktop and it might be better to modify Windows so that they behaved more like Unix.  Windows scripting host now permits the level of scripting comparable to Unix in command line environment and vastly superior to Unix on application level (for example, all MS Office component are scriptable and that alone makes a tremendous difference.)

The same approach is even more appealing for the large corporate environment.  Anybody who thinks that the cost of Linux in a large corporate environment is less that cost of Windows needs to read Red Hat EOL policy several times before going to bed ;-). Even for servers Red Hat with their 12 month EOL thing might well be more costly then Windows. Let's look into two scenarios:

• Scenario 1 - using the "Standard" Red Hat releases. Usually a minimum of  6 month or so need to pass from when a new release is produced before you start rolling it out on low-risk production servers, and perhaps 12 month before you have a complete rollout of an OS. This is an even longer process if you have custom code then it also needs to be tested in a new environment. This means, in reality, a 1 year RHN subscription, on a given box, will only be of use in a production environment for less then 6 months or will not be used at all. Assuming that 8.0 is rolled out here for, say, March 1st, 2003  and 8.1 were to come out on the same day, that would mean that version 8.1 can be rolled out no earlier then August. Now, I'm down to a six month validity period before they EOL (in this case, March 1st 2004). Essentially that means that you have no choice but to upgrade every six months, or go through periods where servers require security patches from a different vendor. Does this create a better TCO than Windows ?  

• Scenario 1 - using the "Advanced server".  Advanced server saves the problem with EOL outlined above. But it is not cheaper then Windows and after calculation in all the costs might prove to be more expensive. Windows 2000 has been out since late 1999 and is extremely stable. If you purchased it then, complete with CAL's and the like, that was your only capital cost. At this point, you would now have had the server for what is approaching 4 years, and won't be required to upgrade (by being EOL'd) for another 3-4 years. That's a very good eight year investment. Alternatively, one would have spent $800 x 8, and have performed 2, possibly 3 mandatory RH upgrades in the same time period. Here RH is considerably more expensive even without mandatory upgrades; the latter might double your costs.

Outside of academic environment and departmental server Linux can be even more questionable proposition. For example, I do not consider Linux to be a superior or cheaper desktop.  It has its advantages as a development machine, but that' about it and with the current hardware costs it's better to have both. The desktop cost turns out to be determined not by the initial cost of OS, but by the cost of hardware as well as availability and quality of  drivers and software applications. The total cost of ownership of  OS it is typically less than 10% of the cost of a typical desktop.  If you work on cheap hardware or use some fancy hardware often the only drivers available are Windows drivers. And  any reasonable specialist needs to adapt to this situation.  Free applications are available both for Windows and Linux, but I would say that all major free applications for Linux are available for Windows too (Star Office, Apache, Perl, PHP, Python, etc) and can reduce the cost of ownership as effectively or more effectively then in Linux.   Reverse is not true.

Also the cost of keeping the basic Windows-office environment are often exaggerated, especially in case of large corporations. For a large corporation Windows environment is as close to free as one can get ;-). Let's assume the following schedule the presupposes upgrades once in every three years) and no corporate discounts:

• 1993: upgrade to Win3.1 ($100) and MSOffice $500.                                                    $600
• 1996: Upgrade Win95 ($100). Upgrade MSOffice 95  ($200)                                        $300
• 1999: Upgrade to Windows NT($100). Upgrade to Office 97  ($200)                            $300
• 2002: Upgrade to W2K($100) Upgrade MSOffice to Office 2000 ($200)                      $300
• 2005: Upgrade to Windows 2003 ($100) Upgrade office to Office 2003 ($100)            $200

That means that organization might spend on basic Microsoft software just $1700 per employee per twelve years or less then $200 per year employee per year. Or less then a dollar a day. In the USA that's approximately the same amount per employee that is spent for coffee :-).  Federica Troni, a senior Gartner analyst, noted the following exaggarations or "myths" as he put it (text is blue is mine --nnb):

1. Linux will be cheaper than Windows because StarOffice can be used instead of MS Office. Not true: Star Office is easier to use in Windows environment then in Linux environment. It's actually better suited to the Windows environment...
   
   
2. Linux is free. Partially true (if you use downloaded copy, but in this case you have no support), but does not matter: the main part of the cost of the ownership is not the initial cost of the OS, anyway.
   
3. No forced upgrades. Not true, especially if you use Red Hat. They are extremly greedy and badly want your money ;-). Moreover Linux kernel is still pretty raw and compatibility can be broken pretty easily if you stay with, say, five-seven years old version (equivalent of Windows NT or Windows 95-98).
   
   
4. Linux will require significantly less labor to manage. Not true, as Red Hat and other distributions are very complicated and tricky to manage. In some areas even more then Windows. Just try to get all the keys of Midnight Commander work on RH9 and compare with FAR on Windows that works OK out of the box ;-).
   
   
5. Linux will have a lower total cost of ownership than Windows because of available management tools. Not true as Linux GUI tools are very rudimentary and some are counterproductive in comparison with their classic command line alternatives.
   
   
6. Hardware will be able to be kept longer if Linux is used or holder hardware can be used. Not true but also does not matter as hardware does not usually survive for more then seven years both in large corporations and at home, anyway.
   
   
7. Applications will be cheap or free. Not true. While set of free tools for Linux is acceptable it is far from the cutting edge. Top editors, debuggers, etc actually can be and often are more expensive.
   
   
8. Transferable skills. Only from other Unixes. 

Why are these myths? Federica produced a table which outlines the total cost of ownership, based on a 2,500 user environment with a mixed user base:

• 1% high performance users,

• 20% knowledge workers,

• 74% structured task workers

• 5% data entry workers.

Even if you take our primitive calculation of the cost of ownership of the Microsoft Office, per person, from 1991 to 2001, and then multiply them by 10K users then the cost of ownership for large organization is  just $3 million dollars per decade. For a billion dollar corporation this is non-essential cost, almost a rounding error. 

Fortune 500 companies usually can negotiate much better deals with Microsoft and believe me that for such a company the cost of upgrade of to, say, Office 2000 will be much less than $200 per user. Often more then 50% less. Actually this difference does not matter much because the real cost is slightly more if you do not have "unlimited" licenses because you need to spend money/time/resources on getting contract, tracking licenses including a non-trivial controls which software goes with which computer. In a large corporation tracking licenses might actually represent more than $100K a year overhead (one person plus some equipment and software), and thus became a major part of overhead. that's why for anybody with, say, 20K user licenses or more it might make sense to negotiate "unlimited" license. 

Of course you need to be careful with Microsoft and when MS.Blaster, Mymail or SoBig hit, I saw several corporations that were crippled for a day or so. That's a lot of money. And that's probably doubles the real cost ownership. But this is just because Windows is a dominant OS, not because similar worms cannot be successful in Linux. Linux in corporate environment is as unpatched (or even more unpatched) then Windows.

There is not free lunch. If you use Linux you need to weigh this against the cost of migrating, retraining and cost of support of open source environment as well as cost of securing the Linux box. I think that for a large corporation just retraining costs alone will be more than  $3 million for 10K users. That means that you did not get any substantial financial advantages for rolling out a pure Linux desktop in a sizable organization for at least a decade.

Moreover, relatively few IS people have good working knowledge or expertise with Linux. Retraining specialists from Windows to Linux is very expensive and takes time. In addition, relatively few organizations are running Linux, overall, so the pool of expertise is limited. But you can cut this costs by implementing Unix utilities and several other parts of Unix environment (shell, gcc, Apache, Perl, etc) in Windows environment as the first step of such migration.

Over time, TCO of supporting a Linux desktop might become gradually less as Linux penetrates deeper into the market. Students often learning it as the major OS because the financial barrier to entry is so low. As the number of experts increase, the TCO will decrease. Meanwhile you can do a lot to make Windows more productive by replacing some weak tools and adding Unix ports of the best Linux software. It's still free...

There are four major Unixification suits:

• ***** uwin. UWIN is the most complete solution that for some reason was not adopted by Microsoft, so it is now in some disadvantage. It includes an X11R6 X-server (absent in SFU 3.5 ) as well as ssh and scp commands (also absent from SFU 3.5).  The KornShell, (ksh93) can be started by double clicking the icon labeled "ksh for Windows NT" in the UWIN program group. In addition The ksh runs in a console window, just like the MS-DOS command shell. Once ksh is running, all of the UNIX utilities can be executed. In addition, ksh can execute native Windows applications. The UWIN console provides an emulation of the VT100 terminal so that programs that use the curses library should work. All the environment variables of Windows that have been initialized when ksh has been started can be accessed from ksh. Some variables, such as PATH, which are understood by both Windows and UNIX utilities, but which use different formats, are converted to UNIX formats when executing UNIX utilities, and converted back when executing Windows utilities. The environment variable DOSPATHVARS can be set to the names of additional variables that get converted to and from native path formats.

  This is definitely much better implementation then Cygwin, but Cygwin is closer to standard an enjoys open source momentum. Whether it's better to use it depends on other open source programs that you want to use.
   

• ???? Windows Services for UNIX (SFU) Version 3.5 Beta. SFU is based on a product that predates Cygwin by a number of years. It was formerly called OpenNT, and later,  Interix  Microsoft then bought it and called it SFU. It has been around for quite a number of years. Unlike Cygwin, it is an actual subsystem of the operating system (similar to the OS/2 and posix subsystems that used to be included by default in the NT OS's). As David McBride noted this is an important architectural difference. While Cygwin is built on top of the Win32 APIs on top of the Windows kernel, SFU is built straight on top of the kernel; Win32 API variances (which have caused headaches for the Cygwin implementors) does not make any effect of SFU. Moreover, the core state information (such as process listings) also come straight from the Windows kernel. That makes possible to send a SIGSTOP or a SIGKILL to Word.exe (a Win32 app) from the SFU universe.  As well as NFS mounting and export capabilities, SFU also supports NIS and can do various user mappings between the Windows and Unix worlds. Beware the default password set for some of these options: no service that requires a password for security should be enabled by default with a standard initial pass phrase.

  SFU 3.5 while far from being perfect is free. Technical support for SFU is provided by Microsoft Corporation support info  Part of the package looks like repackaged and recompiled with Visual C++

  • Interop Store Source code for SFU 3.0/Interix Utilities   $20       
    CD containing the source code for the SFU 3.0 base utilities: diff, sdiff, bc, dc, cpio, gzip, gunzip, gawk, patch, csplit, nl, strings, rpm, and SDK utilities/libraries ld.so, gcc, gdb, g++, g77, gasp, objcopy, ld, as, ar, nm, size, strip, ci, co, diff3 rcs, rlog, and ident.
    
    Notes:
    1. The source code is provided under the terms of the GNU General Public License, a copy of which accompanies the source code.
    2. If you require source code for Interix 2.2 utilities (i.e. not SFU 3.0), please request it in the "Special Instructions box" on the order form.

  See also:
   

  • Interop X Server 8.0 for SFU/Interix An important tool in the migration of a UNIX X-Windows application to the Microsoft Windows platform is a reliable X-11 Server.  Interop Systems Inc. and Hummingbird Communications Ltd. have teamed up to offer version 8.0 of Hummingbird's world-leading X Server technology optimized for Microsoft Services for UNIX / Interix technology. For technical specifications, please view the Exceed Data Sheet at Hummingbird.com.

• **** Cygwin pretty basic GNU software suit, popular but technically inferior to SFU and, especially, UWIN.    Mostly straightforward ports of GNU utilities ( none is impressive: for example compare bash port with the  ksh93 port);  however, the Windows ports sometimes lag the Unix versions in stability and, necessarily, in features: some parts of Unix API are not mapped.   A big plus is the availability of complier (gcc).
   

• *** Morris Kern Utlities. Pretty average, much below Uwin and SFU. Rather expensive. Restrictive licensing.
   

Last modified: February 28, 2008