Kenneth Harker's Linux on Laptops and Werner Heuser's Linux Laptop HOWTO guided me in choosing this laptop and getting Linux to work on it.
The ProStar 3500 won't boot unless there is an MSDOS filesystem present on the harddisk. I made a 7MB MSDOS partition. After that, LILO works as expected.
RedHat 6.1 installed on this notebook without trouble from the CDROM drive. The pcmcia stuff recognized the AT2800TX network card, but was unable to initialize it. The X server that ships with RedHat worked just fine. There were minor problems when doing suspend/resume to disk from within an X session without switching to a text console first.
The Python install script for RedHat seems very unreliable. If I make a mistake and then use the backup feature, it will surely get confused. I tried to install RedHat from a DOS partition on the local hard disk, but in this case, even if I didn't make mistakes the script would fail. Perhaps I was doing something wrong. Anyway, RedHat 6.1 can be installed just fine from the CDROM.
Next, I tried Slackware 7.0. This installed from my DOS partition just fine, but at the login prompt I got the message
crypt: operation not supportedwhenever a username was entered whose account didn't exist. A message like this should never appear. In my opinion, this is a security problem. I tried again to get the AT2800TX network card working, trying all the different tulip drivers from Donald Becker, but didn't have any success.
Finally, I loaded a highly customized version of Slackware 3.0 that I had running on my 486 system. This included kernel version 1.2.13 with ELF patches among many others. This old kernel worked just fine, but of course there was no cardbus support or power management. Nevertheless, it was familiar, and after compiling the latest version of XFree86, I began using the computer for real work.
I recently downloaded version 2.2.14 of the Linux kernel and installed that. After solving the pty problem and a few other things, I have this kernel running on top of my Slackware 3.0 distribution. The suspend/resume functions to RAM work fine from within an X window. I think there are still a slight problems using suspend to disk without switching to a text console first. I have still not managed to get the ethernet card working, therefore I am using PPP with a null modem cable attached to the 486 system to connect my notebook to the network. So the serial port works just fine. I've also tested this port with a serial mouse.
If anyone reading this has successfully used the AT2800TX ethernet card with Linux on any notebook computer I would like to hear from you. Please send me email at ejolson at math.uci.edu. I've heard from one person who also had no success with the AT2800 and a different notebook computer. I think the problem is not with the ProStar 3500 but with the network card. The ability to recognize this card has recently been removed from the PCMCIA package. However, it is still listed among the supported cards in the documentation.
In a final attempt to get the AT2800 network card to work, I reinstalled RedHat 6.1 and tried many different things, including the 2.2.14 kernel, various tulip drivers, and version 3.1.10 of the PCMCIA package. I have now purchased a SohoWARE NE2000 clone which works just fine.
After installing RedHat 6.1 the sequence of commands
mkdir old; mv * oldsometimes actually tries to move the old directory inside of the old directory. This is insane, and though not a disaster, is rather annoying.
This machine weighs more than 7 pounds. The main reason I purchased such a big heavy machine was that I wanted to run X using 1024x768 resolution on the 14.1" display. My colleagues who have purchased the Sony Vaio are always remarking, "Gee, it sure is thick."
I don't mind the thickness, but I wish the plastic frame housing the LCD screen were a bit more rigid, and I worry that the metal plate to which the keyboard is mounted will touch the circuit board underneath if I press a key too hard. Also, there is no protective cover for the ports on the back of the system. Therefore, it may be happen that a random metal object will come in contact with the docking station port. I have attached a small piece of electrical tape to prevent this.
In practice, the machine seems durable enough.
The keyboard is fine. My main difficulties are in finding the delete and the backspace keys without looking. They are the same size as the other keys and not in very distinctive places. There is a picture of the keyboard on ProStar's web site so you can see what the trouble is. Bryan J. Ischo has mentioned this same problem about the ProStar 3203.
Occasionally I type a key and it doesn't appear on the screen. This may be a mechanical problem, or it may be a lost interrupt. In particular, querying /proc/apm to determine the battery level turns off interrupts for some time. There is also a noticeable delay when a sleeping hard disk spins up to write the bash history or the vi recovery log. These are expected difficulties.
After about three months of use the keyboard seems to work a little better than originally. I am perhaps getting more used to the layout. This is particularly important as the lettering is already starting to wear off of the keytops.
The touchpad works great. There are only two mouse buttons and they are a bit far apart for my dexterity. Therefore, I set the Emulate3Timeout in XF86Config to 150ms and ClickTime in .fvwmrc to 200ms and now my clumsiness doesn't matter as much. I had earlier set this to 250ms to solve the same problem, but this higher value seems to trigger a bug that makes double clicking unreliable in newer kernels.
I am currently running XFree86 Version 3.3.5 with the XF86Mach_64 driver. The video chip is detected as a Mach64 Rage LT Pro rev 220. Some earlier versions of XFree86 do not support this chip. The LCD looks great in 24bit color at 1024x768 resolution and there are no bad pixels that I've found. I haven't tested the external monitor connection or the S-video TV output.
This machine has the ESS Maestro sound chip. There is a sound driver included in 2.2.14 for this hardware. It has improved greatly in the last few months. There are supposedly some inefficiencies still in the recording mode related to interleaving the pcm data for the left and right channels, however playback functions well. Furthermore, there is support for up to four /dev/dsp devices.
Depending on the BIOS setting the suspend button saves to either RAM or Disk. Suspend to disk is unbelievably slow. I deleted the suspend partition and am using suspend to RAM when needed. I will eventually try Linux Software Suspend which reportedly uses the swap partition to quickly suspend the system to disk. I recently tried the software suspend patches for the 2.2.14 kernel. It suspends very fast, but crashes my machine about half of the time upon resuming. The symptom is that init gets stuck and then the system is quite confused. This happens with either the patched init or with the original one.
Battery status is reported with /proc/apm. Each read of this file creates a noticible delay as the system turns off interrupts to run the BIOS routines. It is probably a bad idea to allow remote users to read this file since the command
while cat /proc/apm ; do : ; donewill bring the system to nearly a halt.
I get more than two hours of battery life with light use. Under heavy use the CPU fan turns on and a lot more power is used. When ten minutes of power are left, the notebook starts frantically beeping. The volume controls will not turn this down. However, the whole thing can be disabled from the BIOS. I rather like it, though, so I've kept the feature enabled. It is quite exciting.
I get about 500 bogomips. Slightly more meaningfull is the result of John D. McCalpin's STREAM: Sustainable Memory Bandwidth in High Performance Computers.
gcc -O3 stream_d.c second_cpu.c -funroll-loops Function Rate (MB/s) RMS time Min time Max time Copy: 200.0000 0.2460 0.2400 0.2500 Scale: 200.0000 0.2451 0.2400 0.2500 Add: 232.2581 0.3100 0.3100 0.3100 Triad: 232.2581 0.3100 0.3100 0.3100
I have also run a fluid dynamics code. This showed that the ProStar 3500 gives about two-thirds the performance of a single node on an Origin 2000.
f77 -O3 -mips4 -64 (MIPSpro Compilers: Version 7.30)
g77 -O3 -fomit-frame-pointer (GNU Fortran 0.5.25 19991024)
real user system cpu
Origin 2000 (1 node) 26.622 26.471 0.063 300Mhz R12000
ProStar 3500 39.77 39.42 0.05 500Mhz Celeron
I plan eventually to check the Xserver performance and the disk drives. At an intuitive level, everything seems very quick.
I am satisfied with this notebook and its ability to run Linux.