[Software Systems Architecture] the windows

about archives notes tidbits (RSS) music vidconf.net

Technologists.com

home > tidbits2007 > the windows

Looking out the windows

(1965) "Same old places and the same old songs...
It's the singer, not the song." - Jagger/Richards
(1929) "You may forget the singer, but don't forget this song." - A.P. Carter

January 2, 2007 -- While the "real world" mourns President Ford and ponders the past, present, and future of our planet, many in cyberspace find now the time to ponder hardware, software, and Internet platforms. Questions such as the following seem to be asked now, especially:

Should I buy now or wait for faster/better/cheaper?
Should I upgrade to VISTA and/or Office 2007?
Should I switch from Windows to Apple or Linux?
What should I do with the old hardware?

These particular questions are mostly asked from the perspective of personal computers -- the server concerns and dynamics are related, but considerably different. I ask them of myself and they are asked of me fairly often. I've spent a fair amount of time the last few weeks trying to reexamine major issues and rethink answers. Today I'll summarize hardware aspects and plan to follow up with windowing system and application software issues later.

My most recent thinking on these questions was spurred last year when an old 486 machine was donated to my church for use in a prayer room. The donors assumed that it would be fast enough for web browsing, for example, for looking up scripture passages. Windows 98 or older seemed unacceptable from a from Windows security perspective. Seriously considering newer Windows seemed unreasonable both in terms of purchase price and performance. There was a perspective that Linux would be faster than Windows and that Firefox on Linux would be a good solution, even on this ancient box. My own experience with Linux on 486s contradicted this, but I found a 400MHz Pentium II (Celeron) machine and 512MB memory for it.

Since ubuntu has earned such a strong reputation for desktop Linux, I gave it a try instead of the more familiar (to me) Fedora distribution. Ignoring performance, I was very pleased. But it took many minutes just to get to a login prompt and minutes more until Firefox was usable. That was very disappointing, but the price was right, so the machine was installed with warning that patience is a virtue. (Once Firefox got going, performance was acceptable.)

The next impetus has been assessing the sufficiency of my "iBook Mini". At first I was using the KVM to switch between XP on the Latitude and OS X on the iBook, but as the iBook seemed more and more adequate, carrying the Latitude between rooms seemed unnecessary. Though I could run XP on a seven-year-old Optiplex, that machine also runs Fedora 6, so trying to get along without Windows seemed an interesting experiment -- getting Fedora to add pieces not freely found in OS X, e.g., an adequate version of VNC. (Though I had previously built TightVNC for OS X, that build was not nearly as usable as the Windows version.)

I use Fedora all the time on old 450MHz Optiplex machines, and Windows XP had seemed pretty sluggish on the old Optiplex, so the notion of using Fedora as well as OS X seemed quite promising. However, most of the time when I use Fedora, it is from the command line, rarely venturing into X Windows. When I started depending on X Windows on the Optiplex, I was dismayed by waiting for applications to start -- the wait seemed much longer than with Windows XP on the very same hardware.

Soon I was immersed in ad hoc benchmarking of every interesting operating system and browser at my disposal, on hardware dating from 2005 back to 1991. (I have Solaris 10 but have not booted it in a couple of years. The other obvious omissions include the BSD flavors of Unix and Opera.) The basic notions are to time how long it takes a particular hardware/software combination to bring up a usable GUI (X Windows, Mac, or Microsoft Windows) and to measure time from power on until a browser is usable. More and more combinations came to mind. When the 1990 notebook that had run Windows 3.1 and Mosaic would not power up, I spent more time than I should admit getting Windows 3.1, Mosaic 2 and Netscape running on the 1991 JAWS -- thus the ∞ allusion in the table. The table is mostly ordered by decreasing hardware/software age.

Machine CPU Clock Memory	O/S	Seconds to GUI	Browser	Seconds to Browser
1991 Dell "JAWS" 486DX 50 MHz 32MB	Windows 3.11	∞? (80)	Mosaic 2 Netscape 3 Netscape 4	90 100 110
	Dell SVR4 2.2	110	(none)
	Windows 95	70	IE 5.5	125
	Windows NT4	130	IE 6 Netscape 4 Firefox 2	160 165 255
	Red Hat 5.2	145	Netscape 4	250
1998 Dell Optiplex GXa Pentium 266 MHz 192MB	Fedora 1	195	Firefox 1.5	230
1999 Dell Optiplex GX1 Pentium II 450 MHz 384MB	Windows 98	55	IE 6	60
	Windows NT4	75	IE 6	80
	ubuntu 6	145	Firefox 1.5	160
1999 Dell Optiplex GX1 Pentium II 450 MHz 512MB	Windows 2000	170	IE 6	225
	Red Hat 9	210	Firefox 1.4	230
	Windows XP SP2	150	IE 7 Firefox 2	165 205
	Fedora 6	300	Firefox 1.5	320
2004 Apple iBook PPC G3 900 MHz 640MB	OS X 10.4	60	Safari Firefox 2	70 105
2004 Apple eMac PPC G4 1517 MHz 512MB	OS X 10.4	40	Safari	45
2004 Sony Vaio Pentium 4 2800 MHz 1536MB	Windows XP SP2	65	IE 7 Firefox 2	100 200
2005 Dell Latitude D510 Pentium M 1733 MHz 512MB	Windows XP SP2	60	IE 7 Firefox 2	120 185
	Fedora 6	90	Firefox 1.5	110
2006 Apple MacBook Core Duo 2000 MHz 2048MB	OS X 10.4	30	Safari Firefox 2	35 50

(Some disclaimers before a few semi-random observations. There was minimal effort to be even-handed in the comparisons and to try to make the timings reproducible. The two timings listed tend to bring out the worst in most platforms -- once a platform is up and running, it is often much faster to close/re-open a browser, for example. Some of the hardware was disadvantaged by what might seem like unusual peripherals and/or multi-boot configurations. My natural tendency is to advocate Unix derivative systems, i.e., to give preference to Linux and OS X. I use and probably will continue to use Windows XP as my preferred GUI. I own shares of MSFT.)

A 486 with "age appropriate" software seems more competitive than expected.
The software bloat phenomenon may be even stronger in Linux than Windows.
The "bloat" seems most noticeable on obsolescent (1999) hardware with today's software (Fedora, ubuntu, Windows XP).
Linux seemed to bog down between Red Hat 5.2 and Fedora 1, a.k.a. "Red Hat 10".
Windows seemed to bog down between NT4 and "NT5", a.k.a. Windows 2000.
Though browsers are harder to characterize, installation times (not shown) seemed dramatically longer with Netscape 4 vs. Netscape 3 and with IE 6 vs. IE 5.
OS X seems the outright winner, the platform of choice, based on these benchmarks.

coda

I haven't been spending all my time resurrecting Windows 3.1 -- it just seems that way. Besides recovering nicely from cataract surgery, I have resolved essentially all of the LDAP problems that discouraged me two years ago and have proceeded a little further with VOIP. More on something next time.