Newsgroups: comp.os.msdos.djgpp
Subject: Re: large variation in run-time
Message-ID: <3551724B.8ADC8D55@CPRO.DLO.NL>
From: Johan van Ooijen <J DOT W DOT vanOOIJEN AT CPRO DOT DLO DOT NL>
Date: Thu, 07 May 1998 09:35:23 +0100
Reply-To: J DOT W DOT vanOOIJEN AT CPRO DOT DLO DOT NL
References: <354DAFD3 DOT 8D277F06 AT CPRO DOT DLO DOT NL> <6iktlt$4j1 AT argo DOT telecom DOT cz>
Organization: CPRO-DLO
NNTP-Posting-Host: 10.194.2.147
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Lines: 58
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk

Radar wrote:
> 
> >duration of 10^6 runs. For instance, two sessions I predict to take 34
> >and 68 hours, respectively (one should be exactly twice as long as the
> >other), and they took 10 and 43 hours. It is fine that the sessions take
> >shorter, but it is strange that the second is not even close to taking
> >twice as long as the other. So this leaves me with the uncertain feeling
> >of the correctness of the results, although they appear to be what I
> >expect.
> 
> Which function do you use for timing ? I had some strange problems with one
> called PCTIMER13. Now I use standard uclock and it looks OK.
>

Because run-time takes hours, I simply get the time and date:

   FILE *out;
   time_t time;
   struct tm *tblock;

   for (...millions...)
   {
   timer= time(NULL);
   tblock= localtime(&timer);
   fprintf(out,"%s",asctime(tblock);
   ....
   }
 
> >whatsoever. The program uses little memory, it does millions of
> >malloc()'s and free()'s, but total used memory doesn't grow (so that is
> >OK) and there is no swapping activity.
> 
> Are you sure that the program doesn't swap to disk ? As I know (but I am not
> an expert) there is trouble to have certainty that you allocate physical
> memory and not only virtual and swap to disk. Your problem could be caused
> be fragmentation of memory and then swapping to disk in the larger
> simulation (??). In protected mode you do not know when you call
> malloc(1000) how much it really allocates, only that it is at least 1000. I
> haven't found any way how to get to know if program swaps to disk inside my
> program.

All memory that is allocated before the iterations is something fixed
between 50 and 100 Kbytes. The millions of malloc()'s and free()'s
concern only less than 1 Kb. I don't see any swapping activity, and I
don't think that swapping is necessary with this amount of memory on a
32 Mbytes machine; as said "go32" (yes the older version 1.12, used with
gcc 2.6.0) reports megabytes of free memory.
 
> I use CWSDPMI server, so I would try run it only with physical memory (there
> is some switch, -s- I think ), disable all swapping and if the program
> doesnt die on short memory, I would be sure that program doesn't swap.
> 
> I work on similar problem, but smaller simulation and measure time so If you
> find something about this problem I would appreciate If you could share your
> experience with GCC & GO32 & simulation.
> 
> Paul
> p_radar AT hotmail DOT com