From: byrne AT katie DOT vnet DOT net (Paul Byrne)
Newsgroups: comp.os.msdos.djgpp
Subject: Re: profiling problems...
Date: 13 Feb 1998 22:57:13 -0500
Organization: Vnet Internet Access, Inc. - Charlotte, NC. (704) 374-0779
Lines: 38
Message-ID: <6c34mp$omd$1@katie.vnet.net>
References: <6c12pa$gf7$1 AT katie DOT vnet DOT net> <34E4609C DOT 8176C9D1 AT LSTM DOT Ruhr-UNI-Bochum DOT De> <34E47EBA DOT B5DF404F AT LSTM DOT Ruhr-UNI-Bochum DOT De>
NNTP-Posting-Host: 166.82.1.7
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk

Thomas Demmer wrote:
>Thomas Demmer wrote:
>> Paul Byrne wrote:
>[...]
>> 
>> > The report from gprof only accounts for about 10 seconds of computation.
>> > I ran it again and let it go for an hour or so, stopped the program
>> > and gprof accounts for 30 seconds.  (Without profiling the program
>> > exits normally in about 5 minutes.)  It does seem to be consuming CPU
>> > time, it's just not doing anything.
>> With profiling on, the run time is about 2-3 times normal speed,
>> so this is normal.
>> 
>[...]
>Seems to be one of my not that good days...
>Rereading and this time understanding your mail...
>A factor of 120 _is_ strange. Do you run this in a 
>Windows DOS box or under plain DOS? The only idea
>I have right now that running under W95 screws up the timer, 
>if you run under DOS, it should behave quite normal. 
>
>What exactly does you program do? Is there a lot of
>swapping? Any networking stuff included?
>Please post how much memory you have available, 
>and how much your program approximately consumes.

I did some more tests and got a factor of over 200 both
in a win95 dos box and in dos mode.  Just a brute
force computation: no swapping or networking and only
a meg or two in use.  The results were correct, it's
just very slow.

The good news is that after getting the patched libc
this dropped to a much more reasonable factor of 3.
Something in those patches seems to have done the trick!

Thanks for your help!
-paul