CS> I've got 64MB RAM on this P54C/100 system with a fast SCSI-2 hard drive 
CS> (Seagate ST12550N "Barracuda") and was using a 4MB HPFS386
cache.  I just 
CS> tried an experiment of increasing it to 8MB.  My project rebuild time 
CS> dropped from about 84 minutes to 53 minutes with the larger cache.  The 
CS> CACHE386 /STATS also showed an increase in the cache read hit rate from 80% 
CS> to 95%.  This wasn't a completely "controlled" experiment
in that the 
CS> source code changed slightly between builds and I might not have all the 
CS> same software running, so I'll have to try this again under more controlled 
CS> circumstances to be sure about the level of improvement.  But it does seem 
CS> that the 8MB cache is getting a much higher read hit rate and does improve 
CS> performance on this type of task.

I'd be interested in any more results you have from this, including
the next step, try 16 MB cache.

I personally had an application (rescan of my messagebase), which
an HPFS386 4 meg cache made take 2 seconds instead of 25 seconds
for a rescan, because 4 meg allowed all the indexes to be stored
in RAM.  It is very silly IMO to say that there is a 2 meg peak
or whatever.  It depends entirely on what you PERSONALLY run.  You
have to try it for yourself.  For all we know, you may be running
a liquid-helium-cooled 80786 experimental chip attached to a 
Cuban-made 20-meg floppy disk with an access time of 5000 msec.  

Oh, and my messagebase-rescan tests were done when I had 24
meg of memory (only 20 now!).  BFN.  Paul.
@EOT:

---