Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!usc!samsung!rex!mandel@vax.anes.tulane.edu
From: mandel@vax.anes.tulane.edu (Jeff E Mandel MD MS)
Newsgroups: comp.sys.mac.programmer
Subject: Re: async write to disk (MPW C)
Message-ID: <5917@rex.cs.tulane.edu>
Date: 30 Jan 91 15:34:35 GMT
References: <1CE00001.a1e4o2@tbomb.ice.com>
Sender: news@rex.cs.tulane.edu
Organization: Tulane University School of Medicine
Lines: 62

In article <1CE00001.a1e4o2@tbomb.ice.com> time@tbomb.ice.com (Tim Endres)
writes:
>
>In article <9341@dog.ee.lbl.gov>, beard@ux5.lbl.gov (Patrick C Beard) writes:
>> In article <1016@ra.MsState.Edu> krn1@ra.msstate.edu (Ken Newton) writes:
>> #I am working on a real-time application to run on a Mac fx that 
>> #requires very fast data collection to disk.
>> 
>> The Macintosh OS is an unsuitable OS if you need asynchronous
>> access to the hard disk.  The SCSI manager can't work asynchronously in
>> its current design.  Thus you can't take advantage of DMA on the IIfx until
>> a later system (later even than system 7).  A new disk driver will be
required,
>> I believe, or a new SCSI manager.
>
>This is true with one exception. Several manufacturers now sell NuBus
>cards that implement SCSI. These are designed in such a way as to
>allow asynchronous I/O and dual I/O. I helped write a driver for just
>such a card that allowed disk writes to occur while data was being
>read from an image card. This worked very well.
>
>I know MicroNet has such a card, I programmed the Golden Triangle
>ASP card, and I think DayStar has a card.
>
>> However, under A/UX, DMA can and is being used for disk access speedup on
>> the IIfx.  Ironically Unix is likely to give you better disk performance,
>> even though you will probably be spending a lot of time swapping programs
>> to and from disk.  A/UX has no SCSI manager, just different device drivers.
>
>Oh so true, however it is straight forward to design A/UX applications
>such that they do not swap and are very efficient. Plus I believe
>you can access more memory under A/UX than under MacOS currently.
>

  You people are missing the point about real-time programming. Real-time data
comes in from outside the Mac, and there will almost always be a way to
generate an interrupt when it arrives. Even when the disk manager grabs the CPU
for several seconds to copy a big file, interrupts get serviced, and if you are
clever, you can continue to sample in real-time. The trick is to control your
disk writing so that it is efficient (write a whole disk block) and use the
Deferred Task Manager to keep your sampling from blocking the interrupts the
Disk Manager needs to complete its work.
  In any event, DMA will probably not solve your problems unless you are doing
something really compute intensive with the acquired data. The bus is blocked
during DMA, so you still can't get your data off the board.
  As far as real-time Unix, good luck. It has been done, but I don't think it
is pretty. The bigger issue for all of this is to what extent the Mac is the
right platform for embedded control. I have done it; we controlled blood
pressure on a number of patients undergoing cardiac surgery with our prototype
test bench for closed loop control without mishaps. I lived in terror, however,
that someone would decide that they absolutely had to spellcheck their doctoral
dissertation or such during a control run. If I ever do something like this
again, I will probably get the Serial NB card and do the guts of it in A/ROSE.
Flame me if you must, but until you have spent two years trying to make the Mac
push the envelope, try to remember your perspective may be limited.

Jeff E Mandel MD MS
Asst Professor of Anesthesiology
Tulane University School of Medicine
New Orleans, LA

Author of MacInfuse, the most expensive program you will never see.