Kenneth Benoit
Imagine having a powerful workstation on your desk that integrates seamlessly with the campus network. Imagine having a computer that can run a long statistical process in the background while you use Netscape and work on a paper you are writing. Imagine that this same computer is serving your web pages and receiving your e-mail, and that you can access it from home or from anywhere in the world using the Internet. Imagine that this is your multi-tasking workstation using a 200-Mhz super-fast processor, with a high-resolution color graphical X-windows user interface. Now grab a tissue and wipe that drool off your TPM and brace yourself for the best part of all: the total cost of this system was only $2,500!
That's because the computer in this reverie is an inexpensive Pentium/200 running the free Linux operating system. In this article I describe Linux, identify its chief advantages (lots) and its main disadvantages (few). What follows is less of a formal review than an unabashed attempt to convince you that Linux is worth using, or at least trying. Linux users are notorious for their zealotry, and rather than attempt to appear objective I will simply make my prejudices plain. I have been using Linux for my main computing needs for over a year now and cannot imagine using anything else--including much more expensive commercial UNIXes. So in what remains I will describe Linux and its hardware requirements, and then identify two different groups of political methodologists and why they might want or not want to use Linux.
Linux is a version of the UNIX operating system--a multitasking, multiuser operating system originally designed for high-powered mainframes and workstations--for systems generally known as ``personal computers.'' Linux comes from an independent source code base, but behaves exactly like UNIX built on licensed code. In technical terms, Linux reimplements the POSIX specification with SYSV and BSD extensions; in practical terms, Linux is UNIX.
One of the very best features of Linux is its price. Believe it or not, Linux is maintained by a worldwide team of volunteer programmers, and distributed for free! Linux was originally written by living legend and computer folk hero Linus Torvalds in 1991 while he was a computer science student at the University of Helsinki.1 In the past several years, Linux has become skyrocketed in popularity and sophistication. Nearly 85% of computers on the world wide web run some version of UNIX, and it is estimated that a huge proportion of those are Linux boxes.
The Linux kernel--the basic operating system itself--is updated on a practically weekly basis by Linus and teams of programmers around the world. The source code is completely open, covered by the GNU Public license which prohibits selling it except for the distribution costs. This has allowed many commercial providers to bundle their own mixtures of Linux and application software, known as ``distributions,'' but has kept the cost to a reasonable figure of between $15 and $100. And even the commerical distributions may be downloaded directly from the Internet for no charge.
Linux is a completely stable operating system. I have run it for over a year using the beta versions of the new 2.x kernel sources until these were declared ``finished,'' and not a single time has my system locked up. I cannot make this claim for any other operating system I have ever used. Furthermore, Linux use and product support has exploded in recent years and shows no sign of slowing down. Drivers, for example, are hardly ever a problem these days, and commercial software increasingly comes in Linux versions. Perhaps once the bleeding edge, and currently the cutting edge, Linux is on its way to becoming a major established player in the operating system world.
Originally written to run on the Intel 386 platform, Linux now runs on a wide variety of architectures. Most common are the ``x86'' Linuxes, running on the Intel, Cyrix, and AMD versions of the 486, Pentium, 6x86, 5x86, and K5 chips. But Linux also runs on the Digital Alpha, the Motorola PowerPC, and Sun's SPARC platforms. Most importantly, Linux runs well on a variety of hardware. It ran fine for me for 9 months on an AMD 486DX/100 machine, until I upgraded to a screaming Cyrix 6x86/P150+. Linux likes 16MB of RAM and 300MB of hard drive space to run comfortably, although I prefer 32MB of RAM and 500MB of hard drive space to run it luxuriously. There are still people, however, who run Linux in terminal mode with 6 to 8 MB of RAM and around 100MB of hard drive space.
Linux now has drivers for all major hardware, including CD-ROM/CD-R drives, tape drives, sound cards, even wierd things like virtual reality helmets. It doesn't like IBM's MCA bus, but works with just about every other system architecture available.
The Linux installation procedure works just as would installing Windows 95 or OS/2 Warp. Modern distributions-such as Red Hat or Caldera make this process very easy. Installation involves creating a boot diskette, restarting your computer, and following answering questions about what kind of hardware you have and which applications you wish installed. This is also one of the primary advantages of Linux over other UNIXes: all of the important UNIX software you cannot live without, such as TEX/LATEX, Emacs, bash, pine, and so on are installed with the operating system and work as soon as you log in for the first time. This includes complicated things like networking, web/ftp/telnet servers, sendmail, and UUCP. Anyone who tells you installing Linux is complicated is either using an old distribution or has never been anything but a Macintosh user. (I was a devoted Mac user for 5 years, so please contain your flames.)
Several X-windows environments--the standard graphical user environment for UNIX systems--are available for Linux, although by far the free version from XFree86 is the most widely used. There are also commercial versions which claim more hardware support and faster speeds, although I have been able to tell no difference. Graphics performance will depend directly on your graphics hardware. I find that an inexpensive accelerated Mach64 (PCI, 2MB) card gives me excellent performance at 1024x768 on a 17-inch monitor. Motif libraries are available commercially (although work is in progress for a free Motif clone); Openlook and several other window managers are available for free. I have been very happy with fvwm, the ``f'' virtual window manager (use your imagination), which includes a paged desktop, a configurable toolbar, pop-up menus, and behaves similarly to Motif. The latest version of fvwm even comes preconfigured to resemble Windows 95, complete with Start Menu and the Recycle Bin, if such be your tastes.
If you are not a UNIX user, you ask, then what are the advantages of using Linux? Most political methodologists probably use one of the Microsoft operating systems and interact with UNIX only for network communications if at all. For example, let's imagine that you commonly use Microsoft Word and Excel, communicate with Eudora e-mail, and need to interface with Netware printers and volumes. Furthermore, you run Stata for your statistical needs and occasionally Gauss. Finally, you use a variety of other DOS/Windows programs that you just can't do without: Quicken, DOOM, and EzI: A(n Easy) Program for Ecological Inference. What benefits would Linux have for you?
Actually, Linux may have several distinct disadvantages that Linux has for you. First, there are lots of DOS and Windows programs that simply will not run under Linux. For example, the DOS version of EzI will not run under Linux; nor will most of the DOS/Windows software on our campus Novell servers. This said, there are ways to get around this problem. One is to use the Caldera distribution of Linux (my personal choice) for which Sunsoft's WABI is available: a Windows Emulator. WABI will run MS Office 4.3, Word 6.0, Excel 5.0, Wordperfect 6.1, Quicken 4.0 Deluxe, MS Encarta and Bookshelf, Eudora mail, and many others. In addition, Caldera's distribution ships with a Netware client written by Novell that allows printing to Netware printer queues and access to Netware volumes. A second possibility is the free DOSEmu DOS emulator which will run many DOS programs in Linux. The Linux team is also working on a free Windows emulator (named ``Wine'') but it is not yet finished.
Another way around the software problem is to keep Windows on your machine and boot up Linux only when you need to. For instance, I keep several different filesystems and operating systems on different partitions of my computer's hard drives, and use OS/2 Boot Manager to let me choose Linux, OS/2, or Windows 95 when I reboot my computer. This way I can develop DOS and Windows software on my Linux machine. Windows, and most other OSes, will run nicely in their own partitions and not even know that Linux is also there. Most other OSes cannot read Linux filesystems, although Linux can read and write DOS, Windows 95, Unixware, SCO, and NFS volumes, and read-only from OS/2 HPFS and Macintosh HFS filesystems. The majority of individual users of Linux keep some sort of dual-boot capability on their machine. In fact, this is how most people get started with Linux: testing it as an additional OS while keeping their original setup intact. This is also a great way to learn UNIX for a very small financial investment.
A final solution to the software problem is to simply to switch to Linux versions of your old software. Wordperfect for Linux is available, although you could also switch to LATEX, in which case you will probably soon want to start using GNU Emacs and AucTEX. Several Linux word processors and spreadsheets also exist that have no Windows versions. Most DOS and Windows statistical packages also have Linux versions (see below). This is even true for many games: DOOM, for instance. Likewise, there are native (and better) versions of Netscape, Adobe Acrobat Reader, RealAudio, MPEG movie players, drawing and painting programs (xfig is an excellent program not available for Windows), and network communications tools. And: the Linux versions of these are free!
A second problem is that UNIX is not for everyone. The power of a real command shell and the ability to customize your own kernel (the operating system software itself) come with their own costs. Many commercial vendors (RedHat, Caldera, and Slackware) provide some technical support, but you'll have to rely for assistance mostly on your own soon-to-grow expertise, publications such as the Linux Journal and the very helpful Linux community on the comp.os.linux.* groups. I installed and configured my own system in an overwhelmingly Windows/Novell environment, but it was not especially hard, and my network administrator was able to assist with the network configuration. As long as you have an ethernet connection and an IP address, you should be able to communicate just fine with your campus network. I use mail on a different UNIX host, since I sometimes (although rarely) turn my system off, but there is no reason why I could not be my own mail-receiving UNIX host. (I admit being tempted by the snob appeal of having my e-mail address be ken@benoit.harvard.edu.)
While installing a good commercial distribution of Linux is no more difficult than installing Windows 95, it does require some technical knowledge, especially when your hardware is non-standard. Also, the power of UNIX generally is also the power to wreak havoc when you don't know what you're doing. One myth, however, is that using UNIX for your personal computing needs requires you to spend lots of time as a ``system administrator.'' This is no more true than it is for any other operating system, unless you are letting multiple users access and use your system. I occasionally upgrade Netscape or add new Emacs .el files to my system, but otherwise never touch my basic system configuration. And since I log in as a regular user each day I have no power to destroy the system setup beyond my personal directories.
So what are the advantages of using UNIX for you? All benefits delineated in the next section apply to those making their first leap to UNIX. In addition, you will find that Linux provides a far more effective use of your computer resources than does Windows 95 or OS/2 Warp (or, if you are one of the three remaining people who use it, DOS). I found both Windows 95 and OS/2 intolerably slow on a 486DX/66 with 16MB of RAM, yet this same system hummed just fine under Linux. And I used the same software on all three platforms: Netscape, LATEX, Emacs, and Gauss. If you are already using Emacs and LATEX for Windows 95, and customizing your command shell with versions of bash and other UNIX goodies, why not just switch to the real thing? If you are a happy user of Wordperfect who uses Windows 95 and has no idea what bash is, then a world of opportunity awaits. You have no idea of the kinds of uses to which you could be putting your largely wasted system.
A different group of political methodologists that might be interested in Linux are those who already use or have just decided to use a UNIX platform. Perhaps you use an HP or Sun workstation at the office, and are looking for a home machine. Or, you are going to purchase a new system for your office and want UNIX but only have a few thousand dollars to spend. Then Linux may be perfect for you.
The advantage of Linux for you is this: it will run nearly everything you want at a much lower cost, and depending on your hardware, at greater speed.
If you already use UNIX, then having a machine that won't run Microsoft Word is not a problem for you. Linux will run all of the standard UNIX platform software you know and love. Indeed, most distributions will automatically install and configure TEX, Emacs, X-windows, all of the GNU utilities, vi, xfig, Ghostscript and Ghostview, your choice of mailer and newsreader, Netscape (depending on your distribution) and much more. This includes programming languages, libraries, and utilities such as C/C++, Fortran, lisp, assembler, Java, BASIC, make, faces, curses, gdb, xgdb, and tcl. All at no additional cost and automatically installed! The GNU C compiler (gcc) is nothing to scoff at, either: the Linux kernel itself is compiled with it. So is Aptech System's Gauss.
So what about commerical statistical packages? This is a key question for political methodologists, and has two answers. First, if you are a single UNIX user in a PC environment, then you will have to rely on the packages available for Linux. But Surprise! the list of available statistics packages probably includes the ones you use on other platforms. Linux versions exist of Gauss, Stata, MATLAB, RATS, TSP, SST, Shazam, Mathematica, and the free programs Ox and SABRE. Most of these packages now use file formats that are binary compatible with versions from other platforms. That is useful for exchanging files with colleagues or students who might use the Windows or DOS versions of these programs, for example.
Although the growing number of packages available for Linux is already impressive, a few notables are not among them. S-plus currently does not run on Linux and there are no plans for porting it. Nor are versions of SAS, SPSS, or Minitab currently available, although the Wabi port is reported to run S-Plus for Windows. But there are ways around this problem.
For the second type of UNIX user, the one connected to a network of other UNIX hosts and having accounts on these systems, it is possible to run other software remotely. If you have an account on any UNIX system that has software you wish to run, you can run them as X clients on your local machine. Your computer supplies the graphical user interface, theirs does the computing, and as long as your network connection is fast you probably won't be able to tell the difference. This is currently how I run S-plus and SAS on my machine from our remote HP network. This is also a good solution for anyone who can persuade his or her university's statistics or economics departments to give them accounts on their UNIX systems.
This strategy is also useful for computing from home, for those who work already in a UNIX environment. Linux can be configured easily with a PPP client to connect to your campus network or other Internet service provider. This permits you to browse the web, read network news, use e-mail, and--as a nice bonus that costs several hundred dollars to set up with Windows--run X-networking clients remotely. There is no less expensive solution to provide your PC hardware with X-terminal capabilities than Linux.
Not many people use Linux in the political methodology community, but may more should consider using this wonderful open operating system. Not only does it do nearly everything that much more expensive commercial UNIX workstations provide, but it costs less and in many cases will be faster. Linux on an Alpha 500MHz platform will provide you with the fastest microcomputer workstation on the planet, for a fraction of the cost of a machine from Sun or HP that even comes close in performance.
Some benchmark results from my own setup confirm that even a mid-range Linux system will perform comparably or better than many commercial workstations. For example, I ran the Gauss speed comparison tests from Stefan Steinhaus to see how my Cyrix 6x86/P150+ ranked. To my delight its overall performance was better than that of the HP 715/100 and the Sun Sparc 5 and Ultra Sparc 170! I also ran a large EI estimation estimation using Gary King's software. It took 41 minutes, 55 seconds on my machine but 48 minutes, 46 seconds on our departments's HP 715/80. I also ran tests in Stata using Richard Tucker's DyadHard software for the years 1816-1900. Creating the dataset took 1 minute, 44 seconds on the HP but only 43 seconds on mine. These tests and my general experience indicates that my modest desktop using Linux offers more computing power than the HP workstation costing thousands more. I only allow myself to dream what these results would look like with a faster processor, say, a Pentium Pro 200 like the one for which Charles Franklin posted benchmark results on Stefan's page...
Where did I put that tissue?
