Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!jarthur!nntp-server.caltech.edu!seismo.gps.caltech.edu!bruce
From: bruce@seismo.gps.caltech.edu (Bruce Worden)
Newsgroups: comp.lang.c
Subject: Re: Crash a RISC machine from user-mode code:
Keywords: Stress
Message-ID: <1990Aug16.051437.21661@laguna.ccsf.caltech.edu>
Date: 16 Aug 90 05:14:37 GMT
References: <1990Aug13.053147.11714@laguna.ccsf.caltech.edu> <481@demott.COM> <1990Aug15.052856.28006@laguna.ccsf.caltech.edu> <17425@haddock.ima.isc.com>
Sender: bruce@seismo.gps.caltech.edu (Bruce Worden)
Organization: California Institute of Technology, CA
Lines: 44

In article <17425@haddock.ima.isc.com> karl@kelp.ima.isc.com (Karl Heuer) writes:
>In article <1990Aug15.052856.28006@laguna.ccsf.caltech.edu> bruce@seismo.gps.caltech.edu (Bruce Worden) writes:
>>In article <481@demott.COM> kdq@demott.COM (Kevin D. Quitt) writes:

>>>....  But when you try to stress the system, it's going to fail.

>>[Disagree.]

>Clarify.  Are you saying that (a) the `crashme' program never causes your
>machine to crash, (b) such a crash does not constitute `failure', or (c) that
>isn't what you count as `stressing the system'?

I am saying (a) that the `crashme' program would have crashed the system 
under SunOS 4.0.3 and earlier (the fact that it no longer does is 
irrelevent save for pointing out that it was an OS bug and not a hardware 
bug), and (b) that such a crash does not constitute `failure' unless it 
significantly detracts from the machine's utility.  By `significant' I 
mean more than a few percent of a given installation's system crashes 
are caused by the bug.  (c) I would claim that heavy use by a number of 
people developing and running applications (finite element and finite 
difference models, processing seismic data, processing GPS data, GUIs, 
et al.) on a networked system with all the trimmings does, in fact,
constitute stressing a system.  If it does not, then his point is still 
not valid because there isn't much more you could do to stress a system, 
short of running it underwater.  Furthermore, (d) we have had no 
significant problem with this bug.  Finally, (e) I have not heard, nor 
has anyone I have talked to about this issue, of anyone, anywhere having
a significant problem with this bug on a Sun.

Conclusion: A system with the described behavior (crashing from user
mode code) while stressed for a significant period of time has not
failed by any reasonable definition of the word.  A conclusion supported,
at least circumstantially, by the strong popularity of the system during 
the period in question.

Frankly Karl, I'm suprised that you did not kill this subject long ago.
I would have if I hadn't been stupid enough to jump into it in the first
place and felt responsible to follow up.  No insult, rather a compliment,
intended.

--------------------------------------------------------------------------
Bruce Worden                            bruce@seismo.gps.caltech.edu
252-21 Seismological Laboratory, Caltech, Pasadena, CA 91125
--------------------------------------------------------------------------