crashes in a Mathematica environment

Mon, 22 Dec 1997 19:42:12 -0500 (EST)

To Whomever May Be Interested,

I am sending this note to the linux-kernel list because I am having a variety
of system crashes and hangs which seem to me (after 28 years around operating
systems internals) to reflect a basic kernel-level problem. Please believe
me that I am NOT confusing the Mathematica kernel with the Linux kernel!

I am currently running RedHat 4.1 Linux (Kernel 2.0.27) on a 200 MHz AMD K-6
with 96 MB of RAM. My problems are a variety of Linux crashes and bizarre
failures when running Mathematica 3.0 and under X. The Mathematica
notebooks which cause the problems involve many hours of cpu time and often
reach execution sizes (for the Mathematica Kernel) of several hundred MB
(working set sizes are typically around 86 - 88 MB). So far I have
experienced the following problems:

1. Linux reboots during notebook evaluation (this invariably happens late
at night when I am asleep). No core files are to be found.

2. Mathematica Kernal core dumps and exits (may be a Mathematica problem).
This has happened twice after 2.5 DAYS of computation!

3. After exitting Mathematica, cannot re-execute the front-end with getting
a core dump. Must reboot Linux to cure this.

4. Mathematica Kernal hung (Status=D) - process cannot be killed, even
during Linus shutdown (which of course hangs).

5. After Mathematica Kernel core dumps, keyboard entry is not mapped
correctly (correct passwords do not work and Ctrl-Alt-Del will not reboot

6. Linux freezes solid (mouse will not track, etc.)

All but #5 have occurred with Mathematica 3.0. So far only 2 and 5 have
occurred with

The really curious thing is that for a given version of the notebook, the
crash (or hang) always occurs at the same place, and is exactly reproducible.
A small change in the notebook (ie., a different set of steps during a
simplification) will change the behavior, again in a well-defined and
reproducible way.

The computations I am working on are the calculation of scalar invariants
for the Kerr-Newman metric in General Relativity (if you're interested).
They often involve approx. 32000 terms of varying complexity.

Note that I have 4 133 MB swap partitions configured (one primary and
three in the extended partition).

In an attempt to obtain a more stable environment, I am considering
upgrading the kernal to 2.0.32 (I am now downloading the rpms for this).

My questions are two:

1. Has anybody heard of anything like this?

2. Is it safe to upgrade the kernal from 2.0.27 to 2.0.32 WITHOUT doing
the full upgrade to RedHat 5.0?

Thank you for any help you can give me - This is driving me nuts!