🚀 go-pugleaf

In article <18660@cup.portal.com>, Greg_Mark_Finnegan@cup.portal.com writes:
> Jonathan Garber's quote ("...there's only one way to do virtual memory...")
> bothers me a bit.  The Connectix VM scheme limits you to 8Meg period. Apple's
> method lets you go to 14Meg by mucking with NuBus (in 24 bit mode) and gobs
> more in 32 bit mode.
> 
> Sounds like 2 ways to do virtual memory to me (and probably a judge).

First of all, the approaches will be similar in many respects simply because
they both use the same hardware, i.e., a Motorola PMMU in a Macintosh.  That's
not what's patentable.

From everything I've seen on both of these schemes (which isn't any more
than most of the rest of the group :-)), there seem to be some major
differences in how they are implemented.  I think these differences are
simply the result of Connectix being a third-party developer, and Apple
being, well, Apple...

From the descriptions that have appeared here and in the trade rags, Virtual
is a marvelously clever piece of software, but it looks like the authors
have attempted to do as little messing about as possible with the lower
levels of the system, which means that the pager sits (more or less) between
the application and the toolbox.  Since they are using the standard SCSI
manager, they can't handle page faults during I/O operations.  They get around
this by preflighting disk accesses in order to insure that all of the buffers
are paged in before the operation starts.  This does in fact seem to work,
although it causes problems for scanners and other devices that talk to
the SCSI manager directly instead of via the file manager.  I'll call this
method "second guessing the applications," and I suspect that this is what
the patent application covers.

Apple seems to be taking another approach, namely rewriting the SCSI manager
to be reentrant, so that page faults can be serviced even when other SCSI
operations are pending.  This localizes most of the patching to one area,
but it's something I think only Apple can do effectively, since they are
in control of what the standard system software consists of, and I don't see
how it would infringe on Virtual's stuff, since it removes the need for
most of the cleverness involved in using the current SCSI manager.

Both schemes involve tradeoffs:  Virtual involves lots of little patches
all over the I/O system, while Apple's VM involves completely replacing
one OS manager and leaving the rest of the I/O system more or less alone.

Apple's scheme involves a fair amount of cleverness, too, with the idea of
remapping the NuBus to make room for more memory in a 24-bit addressed
system.  Once you see it, it seems obvious, but I still think it was
pretty ingenious on the part of whoever thought it up in the first place.

--
Amanda Walker <amanda@intercon.UUCP>
InterCon Systems Corporation