Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!usc!apple!mattd
From: mattd@Apple.COM (Matt Deatherage)
Newsgroups: comp.sys.apple2
Subject: Re: RamFAST SCSI Card in ProDOS 8
Message-ID: <54485@apple.Apple.COM>
Date: 30 Jun 91 17:18:59 GMT
References: <1991Jun25.070143.9288@clark.edu>
Organization: Apple Computer Inc., Cupertino, CA
Lines: 47

In article <1991Jun25.070143.9288@clark.edu> andy@pro-palmtree.cts.com (Andy Stein) writes:
>In-Reply-To: message from jb10320@uxa.cso.uiuc.edu
>
>    Would GS/OS be faster with the RamFast SCSI Card if GS/OS's cache were 
>disabled from the Control Panel?

An interesting question.  I'll try to keep the hopefully interesting answer
short.

First, it's entirely theoretical, because you _can't_ disable the GS/OS cache
from the Control Panel.  Even when you turn it to 0K, GS/OS keeps a minimum
of 16K cache for it's own use.  More about this mechanism can be found in my
article on the subject in _develop_ #2, the text of which has been on all the
Developer CDs (and _develop_ CDs) since then.

Second, it's possible that it would be a little faster, but it's hard to
tell.  Most of the benefit we see from the cache in general GS/OS usage
comes from the FSTs (well, "FST" since we're probably not talking about
High Sierra here) which manage to cache "system" blocks automatically.  Things
such as directories and bit-maps which the FST is likely to need fairly often
are cached, because when the FST asks the driver to read the directory/bitmap
information, it requests that the driver cache the block.

All caching is done at the driver level -- all other parts of the system merely
request that drivers cache blocks if they can.  The driver may always refuse.
The 5.25" driver always refuses, for example, because it has problems 
determining when disks are switched (and therefore, when to flush the cache).

For not caching to be faster than caching, the driver in question would always
have to be able to return the information from the device faster than it would
take to locate the block in the cache bucket chain and use 65816 instructions
to move it to the proper location.  For a RAMFast, this might be possible if
the cache on the card can DMA any block on the disk into memory without using
any seeks.  If the request is likely to cause a seek, though, the cache will
likely be faster.

BTW, since caching is done at the driver level, the RAMFast people could
fairly easily build a driver that doesn't use caching and gauge the performance
difference.  (Presuming, of course, that their existing driver does use the
GS/OS cache -- adding it would be harder than removing what's already there.)
-- 
============================================================================
Matt Deatherage, Developer Technical  | The opinions expressed herein are
Support, Apple Computer, Inc.         | not those of Apple Computer, and
Personal mail only, please.  Thanks.  | shame on you for thinking otherwise.
^^^^^^^^ Technical questions are not personal. Please post them instead.
============================================================================