On Wednesday 25 Feb 2004 8:29 am, David Matthews wrote:
On Tuesday, Feb 24, 2004, at 21:34 Europe/London, Rob Arthan wrote:
Every time you run poly, it does a memory access that causes a memory fault, which it then handles and recovers from - a perfectly reasonable way to do its memory management, I would say.
Unfortunately, on Mac OS X, the memory fault triggers a crash log to be generated, and this causes quite an overhead (about 30% of the CPU time in my ProofPower build which runs poly a few hundred times). On Mac OS X 10.2, you could turn this off with the Console utility preferences, and that claws back the lost CPU, but you can't do this on Mac OS X 10.3.
Has anyone got a work-around for this? I consider it to be a bug in Mac OS X and would report it to Apple if I had a simple program to demonstrate the problem (but the technique Poly/ML is using is quite a hard one, so that's not trivial).
Rob, I think this is picked up by the CrashReporter process. Have you tried running without the CrashReporter? If you edit /etc/hostconfig and add the line CRASHREPORTER=-NO- then CrashReporter will not be started when you boot Mac OS X. I'm still running 10.2 so I don't know whether this will work in 10.3 but it may be worth a try. Does anyone know if CrashReporter does anything useful?
Thanks - that approach does still work on 10.3. Of course it does mean that you won't get all the evidence that might be asked for if some other application does crash. I wonder why they removed the control from the GUI?
I still think it's a bug that it does this at all and will report it to Apple one of these days.
Regards,
Rob.