At last, it is time for the second part of the Techway Endeavour II salvage blogs. The previous post detailed the computer, its contents and its significance. This post looks at getting it running as best as I can.
Testing … 1 … 2 … 1 … 2
The first question is whether the computer would work, at all. It probably hasn’t been plugged in for at least 15 years and electronics are rarely built to last that long. The presence of the Dallas RTC module was already cause for concern, as they integrate a Lithium cell inside to maintain the SRAM and once that depletes, some boards can never boot again. The Pentium first-generation motherboards were all rather famous for being plagued with it, along with some Sun workstation boards.
Another concern was the power supply. While it looks visually fine, switchmode supplies place high demands on their components and stop working for a myriad of reasons. Those old capacitors might not be quite as good as they were in 1994 – would they still be good enough?
I already knew the hard drive was a potential problem. I’ve seen Western Digital Caviar series drives from that era drop like flies in the past few years – my stack of five drives from the 400MB era have all perished within the past two years. As a result, I didn’t want to chance the drive and left it aside to be dealt with separately so as not to waste any chances. After all, knowing if the rest of the machine would even POST would be good enough.
To do things safely, I decided to use my 300W Pure Sine-Wave Inverter and a lab benchtop power supply to create an isolated and limited power mains supply. This way, if the power supply were to fault, I shouldn’t be causing too much damage or blowing any fuses. I hooked up a PS/2 keyboard via a DIN adapter, a serial mouse and my old 22″ LCD monitor to the VGA port and hoped for the best.
I push the power button. A faint buzz comes out. The inverter trips out on overload. I sigh and quickly turn the machine off. Was that it?
I check my connections again. Nothing seems to be amiss. I give it another shot – pushing the button. Again, a faint buzz comes out but the inverter holds stable this time.
A sign of life begins to show as the board awakes from a long slumber.
The BIOS POST messages begin to print and I’m utterly amazed. The machine is alive.
Getting a Health Check
I furiously hit the DEL key to get into the BIOS. Now is the time to see whether it has amnesia, or whether it’s still got its SRAM and retained some of its settings. Taking these down could be vital, so I get the camera ready to take some shots.
The AWARD CMOS setup utility is very familiar, being quite popular. I had an AMD K6-2 300Mhz machine with the same menu setup.
It looks like the Timekeeper battery is not completely dead yet. It still thinks it’s 2016 – a little more than two years slow, so there’s a good chance it is going to fail almost any minute now. It retained the fact that the hard drive was set up in LBA mode, but had a translated geometry of 969/64/63 – dividing the cylinders by four and multiplying the heads by four. This was really only an issue if you were using DOS software using BIOS-based calls, where getting the translation geometry wrong would cause possible data corruption. The expected 32MiB of RAM is detected and a single 1.44MB floppy drive is configured.
For now, I changed the C-drive to None, as I had uninstalled the drive for further work.
Most of the settings here seem to be default – but note the age of the BIOS when it comes to Boot Sequence. I switched it around all its settings – there were only two – C, A (default) or A,C (which I prefer, so I can boot from a floppy).
It was nice to see that the chipset wasn’t entirely devoid of configuration. While you couldn’t adjust the FSB speed here (as this was still a semi-jumpered motherboard), at least you could tweak the RAM timings and wait states if you wanted to squeeze more speed out of it. I set the Onboard IDE Timing to Fastest to hopefully get a little more speed out of it.
The old fashioned power management on the board is also familiar and annoying, which is why I am glad it is disabled. Modern OSes can handle configuring these sensibly, but older OSes did not have inbuilt management so this was a convenient way to do it. But it also meant that you could sit in DOS and the hard disk would spin down or the screen would blank out. If you weren’t expecting it, you could get quite surprised (as I have been).
Finally, the PCI bus has some configurability as well, with manual IRQ set-up and latency times. Could be useful if you are optimising throughput for a video capture card or a high-speed storage array.
Saving and exiting, the POST completes just fine and it found nothing to boot. So I decided to do a RAM test – the next sort of check-up I normally do. Writing out a Memtest86+ floppy, I shoved it in the drive and booted it.
Rather nicely, the test completed with no errors. But do you see that – the cache speed is 85MB/s and the RAM speed is 48MB/s? They say that RAM is fast and cache is fastest – but even microSD cards can offer more sustained throughput than the CPU cache on this computer. That was not something I had expected, but it shows the progress we’ve made since 1995.
How Many Times Do I Have to Ask? A Hard Drive to Get the Data Back
The next step was to hook up the hard drive to my dedicated recovery machine. I was hoping to be able to read out the data into an image if the drive even spun up at all.
I hit the power button and went straight into the BIOS. The drive spun up and it even detected correctly. I had a good feeling … so I booted straight into Debian to begin the ddrescue when bad noises started to appear. Ca-clunk. Clunk.
I keyed in the command anyway and started the recovery in motion. Read errors were constantly being reported and clunking noises being made.
Early on, the drive was still trying to be fully read-out with bad blocks unscraped but patterns were beginning to appear which might suggest some sort of media defect or head issue.
Once fully trimmed and into the retry phase, the error pattern became more apparent, but some errors were also disappearing.
Over time, the errors continued to shrink and it seemed like this might be the side effect of some media damage, head damage, suspension damage, marginality in a head amplifier, contamination by outgassing, etc. Whatever it was, I had a hunch it would not be permanent.
As a result, I did not give up on the drive, and I left it to run recovery as long as it continued to return data at least once every 24 hours and it continued to spin. I wrote and published the first part expecting this process to take a short time … but …
… in all, it required over 12,346 retries and 18.5 days to complete. But it did complete and all the data was returned. All of this was done without ever shutting down the machine even once – just in case the drive would never spin up again. The drive itself wasn’t healthy so I wasn’t going to re-use it. It is instead kept for “ornamental” purposes.
To replace it, I picked up the nearest drive of a similar vintage doing nothing – a Fujitsu MPA3026AT which was a 2.62GB drive – slightly larger than the original. I’ve had good luck with most Fujitsu 3.5″ drives – they’ve given me no trouble even despite their age. I used ddrescue with the –force option to write the image back to the Fujitsu drive so that we could boot the machine.
Starting it Up After a Vital Organ Transplant
After mounting the Fujitsu in the 5.25″ to 3.5″ adapter rails, I noticed the case had a 3.5″ bay just under the power supply. D’oh. How did I miss that?
Anyway, with the drive plugged in, I met a dilemma – the drive wouldn’t spin up. After a quick nose around, I discovered that there was something funny going on with the IDE on the board, so after unplugging the secondary IDE (which I hooked to the CD-ROM as my preference), the drive spun up just fine. I checked cable orientations, swapped cables but that didn’t help the situation. My gut feeling would be pulling the SB AWE32 and twiddling its jumpers might fix it – but I didn’t want to do that just yet.
So with the drive spinning, I thought I was on my merry way until I tried to detect the drive in the BIOS. It came up with some stupid size which didn’t match the drive … so I knew that was not a good sign. As it turns out, and I had also suspected, this BIOS has the 2.1GB barrier, so I configured the drive as 4092/16/63 translated as 1023/128/63 and set it to LBA mode and … drumroll …
… it’s booting! That’s progress!
My excitement was somewhat shortlived, as soon after reaching a garish desktop, the machine rebooted and continued to do so. I was afraid there might be some marginal components causing instability, so I used the F8 menu to get into Safe Mode and then the rebooting stopped. I suspect (as with many Windows 9x installations) poorly written drivers to be the culprit.
After digging around the install, I discovered a trove of software that was improperly installed, but a few potential troublemakers. So I removed Command Antivirus, removed a few stray entries from WIN.INI and SYSTEM.INI, cleaned up AUTOEXEC.BAT and CONFIG.SYS and was able to stabilise the machine at the desktop.
That was what it looked like after I cleaned up some of the mess including a tiled wallpaper and mis-adjusted icon spacing. We can see the user was likely a Chinese user with the installation of NJWIN, there was a Virtual Gameboy Emulator as well, RealPlayer amongst some other software. But that’s still left to explore …
Instead, I wanted to try the slot-loading CD-ROM. Would it work? I dug out an old DOS CD-ROM driver floppy that used VIDE-CDD.SYS as the driver and had a DTR.EXE test program to check the transfer rate.
The good news is that the drive works … the bad news is the PIO transfer mode is very limiting …
Setting the IDE speed to Fastest and running DTR.EXE under Windows gave a good speed boost but still was unable to take full advantage of the 32X rated drive. It seems the CMD0640 IDE controller really is not particularly great … especially with a limited amount of CPU.
There was also another quirk to the machine – that the 3.5″ floppy is identified by Windows and DOS as a 5.25″ regardless of the BIOS setting. I wonder if this is a quirk due to the low CMOS battery, a BIOS bug or something else. I remember my old AMIBIOS 486 machine did have this sort of quirk at one stage which required me to install the drive as B: instead.
Moving into the Modern Era – Virtualisation
While I now know that I have a working computer, it is a bit of a pain to try and demonstrate it using the physical hardware, especially since it outputs VGA analog video. So I decided to try and virtualise it.
My first stop was go to with VMWare Workstation as I’m already using that for other VMs. Creating a new basic VM container and .vhd, I wrote the contents of the drive straight into the .vhd and immediately ran into an issue – the virtualised machine would not boot with an IOS error. As it turns out, this was because the processor is too fast – this patch seemed to cure it just fine.
With that cured, the next problem was that the graphics was very much screwed up. I installed the VMWare Tools including the VMWare SVGA II driver … but …
… even if you have the driver installed, it refuses to work. Part of the reason may well be that the unit was running Windows 95a, an earlier version which the included SVGA driver is not compatible with VMWare. Unfortunately, 16-colour 640×480 did not really suffice for my likings.
Next option was to try VirtualBox, but since I was using VMWare, I didn’t want to cause any conflicts so I gave that a pass.
Instead, I tried DOSBox as I was aware that Windows 95 could be booted on it and it has decent emulation of SoundBlaster audio amongst other things.
mount d: d:\emulation imgmount c d:\imagefile.img -size 512,63,64,969 boot -l c
After I figured out what commands I needed to use to mount the image (shown above) and boot-strap from it, I thought I was home free.
But again, funny business was afoot. Ultimately, I could only stabilise it by setting the CPU emulation to fixed cycles mode with a low value of about 20,000-40,000 cycles to avoid any of this strange business. Then, I had to boot into Safe Mode, fire up regedit and remove HKEY_LOCAL_MACHINE\Enum to force all the hardware to re-enumerate. This allowed the DOSBox emulation to work more properly, but then I suffered graphic corruption. The fix for this is to change from the auto-detected S3 driver to a different one. Once that was done, I was finally “home free”.
The Fruits of our Labour: A Time Capsule
I’ve tried my best to remove any identifying information from the screenshots below. I do know who this machine belonged to and who used the machine with a high level of certainty and it was consistent with the location where the computer was found.
Acrobat Reader 2.1 and 4.0
I don’t ever remember having an Acrobat Reader 2.1 install, so it was interesting to see its splash screen and online help in the form of a PDF file.
Acrobat Reader 4.0 was more traditional with its help PDF, emulating a regular printed manual.
Microsoft Office 97
As expected from a machine at a university, the full Office 97 Professional suite was installed. There were thesis chapters, journal papers, e-mail messages, resumes and other documents strewn about. Below is a sample of one of them –
I wonder if the person became the person they thought they would be after all these years.
RealPlayer 5.0 was installed – apparently an upgrade is available. Rather unfortunate what happened to them.
If you wanted to play .mp3 files, this was one of the official options from the creators of the codec themselves. That being said, with Winamp installed as well on this computer, it seems more likely that would have been used instead.
Xing were a fairly active player in regards to MPEG players and encoders. Not famed for high quality, they tended to focus on speed and so I have used it in the past as well.
Ulead iPhoto Express
I was surprised as I thought that Apple had the iPhoto name first … but apparently not.
RTA Demo Drivers Knowledge Test
This software was the 1990 version and was commonly circulated. I suspect it originally was swiped from a library computer onto some floppies and taken home, as in the old days, we would have to book time at a library to have the chance to practice on their computer.
Various Flash Games
Very common around the period was small Flash games that could be played inside a browser online or downloaded as a self-contained .exe file of about a floppy disk size for local execution. These kinds of novelty software were very widely distributed on floppy between friends, along with joke/gag programs and sometimes trojans. These ones ran just fine even under Windows 10.
I had a thought – what happens if you open an old version Calculator program from the install under modern Windows? Well, it runs, and About is stylized as Windows 10 which is pretty odd but not entirely incorrect. Other system programs call their counterparts (e.g. paintbrush.exe opens up the modern version of Paint).
There were also FORTRAN 90 and BASIC code on the system which appears to have been used to control some data acquisition equipment at some stage.
ICQ Chat Logs
Whenever you dispose a computer, it’s probably a good idea to wipe it. The machine had a lot of software incorrectly uninstalled just by purely deleting the folder in Program Files. Aside from breaking the system in odd ways, it also doesn’t secure the data well – so I was able to determine the three ICQ accounts used and go through their logs.
Most of it is mundane, but it’s interesting to see that the “free porn” trick seems to go back a long way.
The computer had an installation of Netscape Communicator which was also incorrectly uninstalled. However, a file signature search was able to uncover some rather nostalgic parts of the old internet, although sans a lot of the images.
Who can forget the iconic simplicity of the old Hotmail webmail interface?
The fact that Netscape had its own homepage which would be default on every start-up as your “introduction” to the internet. It doesn’t seem like there’s much of a war for homepage buttons as there were in the past.
The Yahoo Search Engine at the time was probably the most popular search engine in Australia – along with a search for western Sydney university before it even existed. Now that’s spooky.
The Trading Post online – apparently a very successful conversion of a paper classifieds into an online classifieds as well. Don’t hear much about them since Gumtree entered the market, but they’re still around today.
Finally, the New South Student gateway, predecessor to the myUNSW system. I like the part where it says “To obtain your Semester 2 2000 results you will need to use NewSouth Student Online or VoiceMark (9385 1999).” I wonder what that would have sounded like.
Further “danger” lies in the e-mail inbox files. Rather interesting to see the tradition of sending “broadcast” e-mail of good news goes back a long way …
Another thing that seems to go back a long way is the whole “we’ll pay you money to spy on you for market research” trick – this one from e-Trends. I wonder what happened to them?
It was a struggle of sorts but most of the hardware in the box was just fine. The hard disk was very much on its last legs, but with persistence, I was rewarded with a full image with no lost data. The clone was bootable, once BIOS barriers were understood, bloatware removed and software misconfigurations corrected.
Virtualising the image proved to be a slight challenge as well, with VMWare being resistant to proper graphics emulation on the Windows 95a install. It turned out that DOSBox was the best alternative, even though it had to be slowed down to ensure correct behaviour.
A stroll through the install was a time capsule of mid 90s/early 2000s memories, with the machine last used in 2003. Old software bought interesting nostalgia. Browser caches, e-mail inboxes and ICQ chat logs were also recovered, showing that things haven’t really changed all that much when it comes to the types of messages they send. It was interesting to see some of the old websites, even if they were missing the images, as some of these sorts of things aren’t preserved by the Wayback Machine.
There may well be another part to this story in the future – when I try to fix the remaining niggles with the floppy, IDE controller secondary port and perhaps upgrade it to run some sort of demonstration. But that probably will still be a while.