Take out memory to speed up kernel build!?

[Ray Shwake]

seanf at sco.COM (Sean Fagan) writes:

>If Tim has a 386DX processor, and the memory card was a 16-bit card, then I
>am not surprised.  What ends up happening, in a lot of cases, is that the
>kernel snarfs up all of the 32-bit "fast" memory, and puts user code into
>the 16-bit "slow" memory ("fast" and "slow" because of number of cycles
>needed to get a full word).  It does this because, as far as it knows, there
>is no difference; all it sees is one large chunk of memory.

>With the 16-bit memory gone, then, of course, the kernel has to make do with
>the lesser amount of memory.  Depending on the size of the kernel, the
>number of various buffer, activity, etc., 2Mb can end up faster than 4Mb.

	Upon reading the report I also mulled over this possibility,
but suspect something else is involved. Given that SCO UNIX (like ISC
UNIX) requires at least 4 MB to get any real work done, even assuming
one can *boot* with only 2 MB compiles will involve such substantial
swapping as to negate any savings from reliance on 32-bit only memory.

	Given the wretched memory upgrade options on my NEC 386/20,
I frequently ran my old ISC 2.0.2 with 2 MB 32-bit and 2 MB 16-bit.
System overhead went up measurably, but compiles still ran more quickly
since I was no longer spending so much time swapping. (My newer 2.2 will
not boot with only 2 MB, and I doubt SCO UNIX does either.)