Path: utzoo!attcan!uunet!decwrl!elroy.jpl.nasa.gov!usc!samsung!dali.cs.montana.edu!milton!uw-beaver!entropy!dataio!shiloh!fnx!del
From: del@fnx.UUCP (Dag Erik Lindberg)
Newsgroups: comp.sys.ibm.pc.hardware
Subject: Re: Added 386 accelerator to XT, hard drive slows down. Why?
Message-ID: <597@fnx.UUCP>
Date: 25 Jul 90 23:14:25 GMT
References: <AVPMT6W@drivax.UUCP>
Reply-To: del@fnx.UUCP (Dag Erik Lindberg)
Organization: I/Ovations    Kirkland, WA
Lines: 26

In article <AVPMT6W@drivax.UUCP> liberato%drivax@uunet.uu.net (Jimmy Liberato) writes:
>When I install a SOTA 386si (366sx) accelerator into various XT
>class machines the ideal hard drive interleave factor (as reported
>by Spinrite) suddenly jumps up, generally from about 4 to 7. 

Typically what happens with these accelerator boards is that they do
not do disk I/O or keyboard I/O directly.  This is so they will work
in a wider variety of machines without compatibility problems.  The
accelerator typically works by taking over the bus from the 8088
(the 8088 is suspended in a hold state).  When keyboard or disk I/O
is required, an I/O request is queued up for the 8088 and the bus is
released.  The 8088 fires up (executing it's side of the driver BIOS
for the accelerator board), processes the I/O request, depositing the
data and status in predetermined locations, and enters HOLD state
again.  The accelerator then starts up again, and moves the data into
it's own, usually on-board, 32 bit, memory.  The additional sector
skew you are seeing is a result of this double CPU swapping for every
I/O request.  This is why accelerator boards work quite well for
spreadsheets and other CPU intensive tasks, but offer little, if any,
performance boost for things like databases.


this CPU swapping.
-- 
del AKA Erik Lindberg                             uunet!pilchuck!fnx!del
                          Who is John Galt?