🚀 go-pugleaf

A couple more hints and observations...

First, in my quick off-the-cuff tests on a 486/33 with 256K write-back
cache, performance looked like this:

   Internal    External    Time (ON/ON = 1.0)

      OFF         OFF            11
      OFF         ON              2.7
      ON          OFF             1.2
      ON          ON              1.0

Remember that the cache problems I've been talking about are relative to
the *external* cache.  An external cache might cache some parts of
memory but not others.  Now observe from the above table: if, in fact,
you are seeing major performance differences due to only parts of memory
being cached, it's quite likely that your *internal* cache is turned
off.  Otherwise it would hide most of the effects of the external cache.
So before you do anything else, *turn on your internal cache*!  Most 486
BIOSes allow you to set this; if not, there's not much you can do at the
moment.

Both MSv4.1 and ODT 2.0 are supposed to turn the internal cache on
automatically, but the code is wrong and it doesn't actually affect the
state that the BIOS left it in.  MSv4.2 will actually correct this
*correctly*... ;-}

Second, if you're having trouble getting all of your memory cached, the
boot program has one trick up its sleeve that may help a bit.  If you
add "/L" to your mem= bootstring, you tell /boot that /unix must be
loaded entirely below 16MB.  This might help you move your kernel out of
slow memory, into fast.  In fact, if you just upgraded from 3.2v2 to
3.2v4, and it's much slower, the *first* thing you should do is try
"mem=/L".  If that seems to fix the problem, you've got some sort of
cache problem which you should research in more detail...

>Bela Lubkin<  Support engineer, The Santa Cruz Operation