Memory usage problem

[Nose Nada]

Hello there!

I'm using Octave 3.0 running on Fedora 9 in a Dell XPS M1530  (x86-64). I
have a problem with memory usage that do not allow me to go on with my work.
Here it goes.

"data" is a binary file containing a matrix and a cell arrays both of
103500x48 elements. The size of this file is 280 MB.

Before starting Octave the whole system is using around 300MB of my 3GB of
RAM
Once I start Octave, this program takes around ~30 MB of RAM

However, after
 
octave:1> D = load('data');

Octave jumped to use up to 1.2 GB of RAM. Remarkably, the memory alocated to
D is just 65MB, which is far smaller than the increase from 30MB to 1.2 GB.

octave:2> whos -l
...
*** local user variables:

  Prot Name            Size                     Bytes  Class
  ==== ====            ====                     =====  ===== 
   rwd D               1x1                   65014358  struct
   rw- __nargin__      1x1                          8  double
...

global variables and functions displayed by whos -l occupy just few bytes.
Then, the allocation of memory to D does not explain the jump from 30MB to
1.2 GB in memory usage by Octave.

Furthermore, after

octave:3> clear D

D is cleared but the usage of RAM by Octave keeps at around 1.2 GB. 
Just when I close Octave the RAM returns to pre-Octave levels.

The situation gets just worse if I load the file as part of a function with
just a few more commands (dir, if,isempty, regexpr). In this case my 3GB of
RAM get exahusted and 300MB of swap are used.

What is going on here. How can I manage that the usage of RAM becomes
coherent with the size of the data I'm loading. At this point I haven't been
able to keep using Octave for my work because the extensive use of sawp
memory, up to 4GB for bigger data files, results in a very poor performance.

I look forward to hear from you.

Regards,

Omar
-- 
View this message in context: 
http://www.nabble.com/Memory-usage-problem-tp18885641p18885641.html
Sent from the Octave - General mailing list archive at Nabble.com.