The X Windows Solution


A Sun 3/60 Workstation in the late 1980s.

I recently did an interview for an educational video about quantum physics. The filmmaker, who needed some cutaway shots to stitch the interview segments together, suggested a shot of me walking through the server room in my building. I complied, gazing with interest and concern as I strolled past the machines. But I felt very uncomfortable about the phoniness of this scene, because I had never been in the server room before and had no idea what I was looking at.

The experience reminded me, though, of the one time in my life when I felt like I was near the cutting edge of the digital revolution, nearly 20 years ago, as the World Wide Web was just emerging as the Next Big Thing.

A radical transformation in scientific communication had begun in 1991, when Paul Ginsparg launched what was then known as xxx.lanl.gov, and is now called arXiv.org. Back in those early days, a broadcast email would announce to the world the latest batch of papers received at the arXiv, and one could request a paper by sending an email with “get [paper number]” in the subject line. After receiving the TeX source file (not LaTeX back then) by return email, you could run “tex [filename]” from the command line of a dumb terminal, which would create the dvi file. Then you would say “dvips [filename]” to create the PostScript file, which you could print and recover from a room down the hall from your office. We all agreed that this system was wonderful compared to what had preceded it — preprints stuffed in Manila envelopes traveling by third class mail.

One day in spring of 1993, my colleague John Schwarz, having recently returned from a visit to Brown University, excitedly reported to me what he had witnessed there. At Brown, no one had to mail xxx.lanl.gov to request a preprint. Instead, the paper would miraculously appear on a computer monitor after just a push of a button on an optical mouse! John suggested we set up a similar system at Caltech, and I enthusiastically agreed. Grabbing the latest preprint and viewing it at your desk without even rising out of your chair? It seemed almost too good to be true!

Except, after a few minutes of sober reflection, I realized with a gulp that the responsibility for achieving this marvelous technological leap would fall upon me. Months of misery ensued, for though I had by that time somehow acquired an undeserved reputation as the “computer savvy” member of the Caltech particle theory faculty, I was actually way out of my depth.

Paul Mende in the early 1990s.

I had about $40,000 to spend and needed to provide a GUI for about 30 users. Options were limited. After a lot of research, I decided to copy the configuration at Brown. The system administrator there was Paul Mende, a very good string theorist and a really great guy, who helped me immeasurably. (Later, Paul left physics to start a very successful hedge fund.)

The solution, which I copied from Paul, was to provide each user with a Sun 3/60 workstation (with 4 MB of RAM) and a 16″ Sony color monitor. By 1993, the Sun 3 was well past its prime, yet many used ones were still available for a reasonable price. Seth Robertson at Columbia had written a neat program called XKernel, which made the Sun 3 think it was an X terminal. The jewel of our system was a spanking new SPARCstation 10 from Sun Microsystems, which was pretty state-of-the-art and capable of serving all those X terminals. The Sun 3s devoted their paltry processing power and memory to displaying the graphics, while the SPARC 10 did everything else.

To avoid being a crybaby I won’t go into all the things that went wrong, many of which involved painful negotiations with sketchy vendors about worn-out used computing equipment, or getting those fussy Sun 3s to do what I wanted them to. My partner in suffering was Ming Lu, a graduate student in our group whose expertise I relied on, and who patiently put up with my many ignorant questions. A major hero of the story, aside from Paul Mende, was (then) Caltech undergrad Steven Fought, who knew everything about Unix and rescued me from disaster multiple times.

In the end there were rewards. On October 1, 1993, the particle theory group server www.theory.caltech.edu joined the burgeoning World Wide Web on planet earth. (The URL is still in use, but the pages have evolved a bit.) Ours may not have been the first publicly available web server on the Caltech campus, but it was among the first. I painstakingly wrote the html code for all the pages myself by hand, though I borrowed freely from the pages at Brown and elsewhere. The Mosaic web browser had just been released that year, setting the stage for the Internet boom. It was fun to feel like an early adopter as the technology surged.

Maybe the decisions I made were not the best possible ones (by a few years later there were much better options), but we got several years of decent use from those battered Sun 3/60s. The X Windows solution was appropriate for that era and the budget we had.

Looking back, though, what I remember most vividly are the days of despair and the sleepless nights when the whole project seemed doomed. In the end, I persevered by getting others like Paul Mende and Steven Fought to help me do what I couldn’t have done on my own.

Often, when one gets into trouble, the best way out is to ask other people for help. I have learned this lesson enough times, but still I forget and have to learn it over again. Why doesn’t it stick?

8 thoughts on “The X Windows Solution

  1. Pingback: » “when one gets into trouble, the best way out is to ask other people for help” Gordon's shares

  2. I remember that, especially seeing my first glimpse of the WWW on John Schwarz’s terminal (was that the Lynx browser?). I guess I just missed your web server debut, as my postdoc ended in ’93. I did go on to build the first version of http://www.cmu.edu/physics/ at Carnegie Mellon.

  3. it is illuminating to see this from the caltex end

    > and one could request a paper by sending an email with

    or by anonymous ftp

    > After receiving the TeX source file (not LaTeX back then) by return email

    latex was a mid ’80s thing, and actually the second article in Aug ’91 (hep-th/9108002) was latex (the first was amstex), and 108 of the 301 submissions made in ’91 were latex (i.e., > 1/3). eventually latex became dominant (though even more remarkably, some people *still* use harvmac, that other 80s thing, since left unsupported).

    > from the command line of a dumb terminal

    the link was to a 1978 vt100, but by 1983 dec had moved to vt220 (and vt420 by 1990), so it’s possible, though not likely, that caltech was still at vt100 in 1993. and many people had already moved to modern graphical user interfaces by then, not only with macs starting in ’84, sun workstations (as at itp, now kitp), but also the 2nd generation NeXTstation had come out in 1991 so many of us had already moved to modern graphical user interfaces (though apparently not at caltech?). the NeXTstation ran what is now known as mac OS X (note that the Mail application still has the same bugs i reported to them in 1991, in what was then known as Mail.app). in ’92 i started using tim b-l’s worldwideweb.app under NeXTstep to access slac-spires, then switched to X-windows Mosaic in ’93. xxx.lanl.gov had a gopher interface set up by joe carlson in ’92, then a web interface starting in ’93.

    > One day in spring of 1993, my colleague John Schwarz, having recently returned from a visit to Brown University, excitedly reported to me what he had witnessed there. At Brown, no one had to mail xxx.lanl.gov to request a preprint. Instead, the paper would miraculously appear on a computer monitor after just a push of a button on an optical mouse

    (John S, allegedly accustomed only to a vt100, really reported back that he’d seen optical mice?)
    paul m was an important early enthusiast, and i think he set up something to automatically retrieve and process the articles, making the postscript available via his local server. eventually he and a few others donated their scripts, none 100% reliable, and we folded them together into the autotex on the main server; then added on the pdf distiller when that became available.

    > John suggested we set up a similar system at Caltech, and I enthusiastically agreed.

    you wouldn’t have needed the autotex, since you could have just pulled things down from mende’s site at brown, so presumably it was just a question of providing the gui’s. workstations at that time were still in the $4k+ range, so that would have been only 10 on your budget, and the decision to go with 30 X terminals wasn’t crazy. (this was about a year before the intel-based pc’s came from below and started surpassing the “workstations” in processing power.)

    > Months of misery ensued, for though I had by that time somehow acquired an undeserved reputation as the “computer savvy” member of the Caltech particle theory faculty,

    wow, with collective expertise like that no wonder you got tenure.

    • John S. might not have mentioned the optical mouse, but the Sun 3 had one and I was quite impressed when I saw it. Our terminals may not have been vt100s, but they were definitely dumb, though I had a NeXTStation at home by 1993 (it was great but the monitor was grayscale and mine did not have a web browser — besides my connection to the Internet from home was through a 2400 baud modem). I wasn’t yet using LaTeX in 1993 — most of us at Caltech were still using the harvmac or phyzzx macros at that time. How I got tenure is an enduring mystery — I was so grateful that I accepted onerous tasks like providing gui’s for 30 people. Thanks for saying my decision was not crazy.

    • I was using jyTeX, developed at Berkeley by Jon Yamron, who later went on to Dragon Systems and worked on voice recognition. Paul, thanks for being gracious enough to support jyTeX for its tiny group of followers.

    • Interesting stories from an interesting commenter, Dr Ginsparg! ;-) Just to be sure, my thumb is the thumb-up.

      Harvmac is a fun macro and a religious cult, too. It’s great to see we’re on the same frequency but I won’t ever forget its cult status. Of course, I’ve met many Harvmac diehards but my adviser Tom Banks may have been the hardest one.

      Still, he wasn’t hard enough so I gradually converted him from Harvmac (and perhaps Phyzzx, another plain TeX macro used by some of his typical co-authors) to LaTeX, too. :-) After 1998, the conversion may have been complete.

      In some sense, his conversion from a hardcore Windows hater to a happy Windows NT user was even easier. :-)

Your thoughts here.

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s