Re: gnuplot in Octave review

[Thomas Ilnseher]

Am Mittwoch, den 11.02.2009, 09:04 -0600 schrieb Jordi Gutiérrez
Hermoso:
> 2009/2/11 Sergei Steshenko <address@hidden>:
> > I once had 200+ 'gnuplot' windows created by just one 'octave' session,
> > cumulative memory consumption of the session + the windows was much more 
> > that 4GB (architectural limit of a 32 bits system), and it was on a
> > 32 bits system.
> 
> I don't understand. Why is it good that the memory usage is much more
> than 4 GB? 
That isn't good, it's just the case when you have 200 windows open *
> How could it be more than 4 GB on a 32 bit system? Wouldn't
> that bring your system to a trash?
No, It was the combined memory consumption of all 200 processes. even if
every process 30 MByte, that will sum up to 6 GB. 

modern x86 processors can access more than 4 gb, you just need to plug
that much ram in there. But You can't give a process access to more than
4 GB easily.
> 
> Moreover, why did you have 200 windows open? 
dunno
> And why would a separate
> process be better to have 200 windows open?
if the combined memory requirement for 200 Windows is > 4 GB, than this
is better, else the system will crash.

*) The executable code (.text) is loaded once into the physical memory, 
and then mapped into the adress space of every of your 200 processes.
Thus, if eg. the .text is 10 MB, and you just sum up the address spaces
(MEM), then you will end up with 2 GB even if it does just occupy 10 MB
physical ram. IF all those 200 wins are generated by one process, then
this will implode to only 10 MB address space (MEM).

Same is true for shared libs (that's the idea behind shared libs, btw).
If a 1 MB lib (with PIC) is mapped into the address spaces of different
processes, and you sum up all the different address space sizes, you
count the same lib over and over again. if the lib has textrels, you are
doomed, but the counting is correct.
> 
> - Jordi G. H.
> _______________________________________________
> Help-octave mailing list
> address@hidden
> https://www-old.cae.wisc.edu/mailman/listinfo/help-octave
> 
-- 
Thomas Ilnseher <address@hidden>