Douglas A. Tutty wrote: ... > So perhaps to add to this entry for the FAQ, something that address this > desire to shrink the kernel to save memory: > > "... For standard i386 old computers with little ram, > recompiling the kernel does not provide enough free memory to > affect what you can then do with that old computer. You are far > better to just add a bit more ram."
much closer to something I'd consider adding. :) > I know that other distros have dropped actual 386 CPUs from their > supported list so that i386 actually needs minimum 486. The reasoning > I've heard is that the amount of memory required is too much for any > remaining actual 386 boxes to actually have. Same thing was done recently with OpenBSD, actually. There are better reasons, however... The big one was the 80386 was a "first generation" 32 bit processor for Intel, and there were a lot of ugly work-arounds in the OpenBSD kernel for 80386 systems that didn't need to be there for 80486 and later systems. Dropping support for the 80386 simplified the kernel code, and as we know, that's a very good thing. There were some practical reasons why you don't want to use OpenBSD on an 80386 system: 1) OpenBSD /requires/ a hardware FPU. The 80386 chip did not have it, you needed to add-on an 80387 Math coprocessor, and a relatively small number of 80386 machines had this. 2) There are things we "just do" today that were big deals back in the 80386 and before days, simple little things like compressing a file. Simply loading an 80386 system with OpenBSD was an all-day affair, due mostly to the time required to uncompress the *.tgz files! 3) IDE disks were not common on 80386 systems. You don't want to try to install OpenBSD on an MFM or ESDI drive. Even what should have been an easy retrofit was complicated by inflexible BIOSs. 4) 16M RAM was almost unheard of back then...and many of the 80386 systems of the day were using different RAM than more modern systems do, so the likelihood that you had an OpenBSD-capable 80386 was very low, and upgrading it to being OpenBSD-capable was cost-prohibitive. When this was done, no one had been sending in 80386 dmesg's for a long time. Even before then...the 80386 code spent some time broken around the 3.3 days..and only a couple people had even noticed (we didn't even realize it wasn't broke machines until we realized that several people were seeing the exact same problem!). > I know that my old PS/2 Model 70-A21 was a 386 with 4 MB Ram (at $1K per > MB) and I think it could take a maximum 16 MB (but my memory from 1988 > is very fuzzy). Where there any 386 boxes that could take 32MB ram, and > do any still exist? oh, most certainly. The VERY FIRST generation of non-IBM-brand 80386 (i.e., Zenith, Compaq, AST, etc.) systems were basically 80286 systems with a faster processor and almost everything else carried over from the 80286 siblings. However, by the time the second generation rolled around, the systems were starting to make use of the 80386 potential (though the OS and apps were still treating the 80386 as a really, really fast 8088, for the most part. I'm most familiar with the Zenith systems, as that's where I was at the time, but the second generation Zenith 80386 systems were capable of 20M on board (supposed to be 32M, but a bug was found with support for actual production 4M 72 pin SIMMs (which didn't even exist when the machine was first shipped!) that limited their use to only the lower four slots, so limit was 20M, though later boards fixed that and were able to use all 32M. I recall no customers complaining about this bug. :) I've got several of these second generation Zenith machines still, one of which was, according to the dmesg log, the last systems to run OpenBSD on an 80386. I also have a no-name clone board which I'd put 8x4M 30 pin SIMMs in for 32M, as well. By the time I had the resources to do this, I'd long got and retired much better machines that were capable of running OpenBSD. I'd actually be surprised if the IBM Model 70 was design limited to 16M, though it is likely there was just no physical way to put more than that in. The PS/2 MCA machines were much more advanced than the ISA-standard machines of the day, though a pain in the butt to work with and incompatable, but I'm pretty sure all 32 bits of address lines made it out to the bus. HOWEVER, the 80386sx was a non-starter for a long time: these machines only had 24 bit address buses, so it had a max of 16M, and being they were "cheap" machines, the actual potential of most of the hardware they were used in was 12M, 8M, or way, way less. I don't know that I have ever seen an 80387SX chip -- kinda bizarre thing, an expensive accelerator for a machine you bought because you didn't need much speed... Nick.
