I upgraded from a 286 to a fast 486 back in the day, I never had a 386 and recently I have been itching to do something useful with a 386.

By no means an “ultimate” 386, I nevertheless aimed for a DX as a SX is in my opinion too crippled. I dug around some old parts I have found years ago and were kept elsewhere and found this nice looking 386DX board with a 33MHz on it. The letters saying ISA-386C looked a lot like the font Asus uses, when looking it up it looked as if I was right. Apparently this is the first ever Asus mainboard. Mine is from May 1991 it seems.

First I tried getting it going but it was deadish, seems as if the memory connectors do not make good contact. I could get it alive sometimes by fiddling with the simms. Eventually I had it going semi repeatably with 4x1MB simms (3 chip variety) in bank 0, as my 9 chip set of 4 was working hit-once-miss-five-times kind of style.

Next headache was getting an IDE adapter going. Apparently my whole pile of super io cards are broken. One even locks the system with a garbage screen when lucky, no post otherwise. I have an ide/floppy only card that says EASTERN (probably parody on western somethings being a name back then) with a faulty fdc, but I vaguely remember its ide worked 20 years back. Dead too, until I flipped its mystery jumper to the other position and suddenly it not only recognised the Caviar 2850 I connected to it, but even started booting from it! It had Red Hat Linux 7.3 installed which I used 17 years ago in a different system for another purpose and shelved the small drive since then. Apparently I was quite hard up for working disks back then 😀

It booted the kernel with 4MB, but panicked soon afterwards with out of memory and no process to kill. I put my 9 chip set of 1MB’s in bank 1 and the 386 finally having tasted some action in more than 2 decades promptly counted them and booted linux!

I had a blast with it the past week. (Or did it have a blast with me?) My plan is to eventually host a specific blog on it, so I started testing out various ideas and options. I would have to write the software myself it seems, as it is only sensible to publish the pages as static html to end users and get some decent performance. Apache 1.3.27 with mod_php 4.1 works and runs not-too-bad with a sane amount (2 max) of workers. Resizing images for my application is important, which I managed to do with netpbm, which is a great piece of software. pnmscalefixed (no fpu) scaled a 4032x3024 image to 800x600 (decode, scale, encode tied together with pipes) in about 20 minutes. No problem. It would need to run as a background worker process but I expected that. It uses no more than 1.5MB RAM doing it.

I played with ssl as well (had to drop in i386 version because of illegal instruction) and benchmarked rsa and blowfish cbc (no aes yet in this version):

sign verify sign/s verify/s
rsa 512 bits 0.4806s 0.0415s 2.1 24.1
rsa 1024 bits 2.6325s 0.1414s 0.4 7.1
rsa 2048 bits 17.2600s 0.5276s 0.1 1.9

type 8 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
blowfish cbc 134.75k 177.59k 184.25k 186.50k 187.79k

Stuffing over 180 kilobytes per second through a cbc with a real 386 is downright impressive.

Networking wise I wanted to use one of my ne clones, but neither of them showed up on any reasonable io address. Strange. Then I remembered reading here that a 3c509 “just works” so I gave a sigh and plugged mine in. It worked. Sort of. Discounting its half duplex nonsense and pio manners, I get around 330 kilobytes per second over it. Meh. Once it hung up with an infinite loop error and I had to power cycle the thing to get it back, but at least it works. Sort of. Pretty shitty thing, confirmed again. Will be looking out for something else. Only other cards around here currently 3c503 8 bit isa, but those are even more brain dead.

I was making space available this morning to install the kernel source so that I can compile a smaller 386 only kernel (the glutty standard one consumes about 2.5MB RAM!) when the WD AC2850 suddenly stopped reading and hung the system. It spins up but sounds different at the end of calibration. I plugged it into my usb to ide bridge to probe it a bit, but every sector now has a read error. It is dead.

I will dig around and make some plans and continue.

Meanwhile, has anyone ever seen such strange connectivity issues with memory? Once it works, it works intensively for hours with no parity errors, but sometimes the system will not post. Re-seating memory or moving simms around gets it going again, which sounds like contact problems, but only manifesting on first power up.