Mathematica 2.0 & Mac sys7
- To: mathgroup <mathgroup at yoda.physics.unc.edu>
- Subject: Mathematica 2.0 & Mac sys7
- From: Roberto Sierra <73557.2101 at compuserve.com>
- Date: 29 Apr 92 04:45:29 EDT
Reply to: Mathematica 2.0 & Mac sys7 -> crashes
> I'm running 2.0.0 on a IIfx under system 7.1.1.1 with 20 Mbytes
> of RAM and a 12 Mbyte partition and Mathematica crashes
> quite frequently.
Actually, the problem you describe may not have anything to do
with Mathematica (per se). I've been using Mathematica frequently
using 80MB virtual partitions to do extensive 100-200 frame
animations (that often eat up 24 hours of CPU time), and I've never
encountered the type of crash you're talking about.
> In a recent MacWeek issue there was an announcement that
> there would be a forthcoming upgrade in Mathematica with
> "improved speed and memory management". This indicates
> that WRI is aware that there is a Mac related problem and
> is taking steps to correct it.
Actually, since I've never had a crash or bomb using MMA in
extensive virtual memory / number crunching tasks, I interpret
the phrase "improved speed and memory management" to mean
"improved speed" (self-explanatory) and "better" memory
management. By "better" I mean that MMA doesn't run out
of memory so often, or that the tasks themselves require less
memory -- but not necessarily a bug fix. Unless MacWeek
specifically described a "bug", there may be no bug present to fix.
I've been quite impressed with Mathematica's solidity in 32-bit
and Virtual Memory environments. I would suggest that the problem
is in the operating system, and there may be a simple solution to
the problem, independent of Mathematica.
There are actually two separate components to what one thinks
of as 'virtual memory'. The first is, of course, the 'Virtual
Memory'
control panel item, which lets you enable and disable swapping of
memory contents to disk. The second is the 32-bit vs. 24-bit
addressing mode selection. On my lowly IIx, 32-bit addressing
is enabled using the Apple/Connectix MODE32 INIT independently
of the virtual memory setting. On newer machines (IIfx's and the
like), the button appears elsewhere, though I forget where, offhand.
It is possible to turn virtual memory on and disable 32-bit
addressing,
in which case the total memory space that can be used is 16MB minus
1MB per NuBus card, regardless of the amount of internal memory
installed. For example, for my Mac with 2 NuBus cards, the largest
swap space I can define using 24-bit addressing is 14MB (16MB-2MB).
There is no limit (it's huge) to the virtual space possible under
32-bit addressing.
Without virtual memory, you can still enable and disable 32-bit
addressing, though 32-bit addressing won't be useful unless you
have more than 8MB of internal RAM. The situation you describe,
in which there is 20MB internal RAM and Mathematica is running
in a 12MB partition, suggests to me that 32-bit addressing is ON
(so that you can see all 20MB, and not just 16MB) and that Virtual
Memory is OFF (since you imply that you bought the RAM to prevent
swapping problems). I'd suggest that the problem may lie in some
32-bit unclean software that you are using -- my experience tells
me that MMA itself is 32-bit clean (actually, I know this to be
true of MMA 1.2, not 100% sure about MMA 2.0), and that the operating
system is 32-bit clean.
Suggestions:
- Try going back to 24-bit addressing -- see if the problem goes
away.
- Disable all INITs/CDEVs and try again.
Has anybody else encountered this type of problem?
If so, Was Virtual Memory enabled?
Was 32-bit addressing enabled?
What type of Mac was it?
Which system version?
Which MMA version?
What other INITs and CDEVs were active?
> i specifically went out and bought the RAM because i was
> experiencing alot of crashes using system 7's virtual memory
> [which is definitely inferior to the Connectix's system 6 virtual
> memory product] but the use of RAM didn't help the problem (in
> fact, the problem is much worse under tuneup 7.1.1.1 than it
> was under 7.0 which makes me think its an Apple problem)
Again, I don't think the problem is in the virtual-memory part of
the operating system. I've never had a problem using MMA under
it, and I would disagree with you that the quality is any different
than the Connectix Virtual INIT product for System 6.0.x (well,
there may be slight performance differences, but considering that
swap time is typically only about 15% of the total time in the
kinds of MMA calculations I do, those performace differences
wouldn't amount to much). I *strongly* suspect the problem
is that MMA crashes when 32-bit addressing is enabled, which
would be the same case regardless of whether you were using
virtual memory or internal memory. Try turning it off (if you
only need a 12MB partition, you don't need 32-bit addressing.
If this doesn't help, look to your INITs, CDEVs, etc. I think
you may find the problem has to do with one of these.
Either way -- good luck!!! I know it can be a nightmare, dealing
with these issues, but I hope you can identify the problem.
FYI -- My system consists of:
Macintosh IIx (oldie but goodie)
80MB internal + 80MB removable Syquest drive
8MB internal memory
System 7.0.1 w/tune-up (and 7.0 prior to that)
MODE32 INIT to accomodate 32-bit addressing
Virtual Memory panel, patched by a little doo-dad called
'VM-Eject' that allows the swap file to be defined
on removable media, like my Syquest 80MB drive.
\\|//
- -
o o
J roberto sierra
O tempered microdesigns
\_/ 73557.2101 at compuserve.com