Re: reshape slowdown?

[mavram]

On Wed, Feb 23, 2005 at 07:30:09PM +0100, Francesco Potorti` wrote:
> >As suggested by P. Kienzle, I got a function which, with
> >increasing n decreases to a plateau value and then starts to
> >rise, first mildly and then pretty steeply. A further increase 
> >of n causes octave to abort the job with a "memory exhausted" 
> >message.
> >In my case the plateau is (roughly) between n=350 and n=1200,
> >the onset of the steep rise at about n=3000, and the breaking
> >point at about 3700.
> 
> If you send me the instructions, I'll run it here.  I did that for the
> instructions sent by Kienzle, and I got this:
> 
> octave> n=15;
> octave> t=zeros(1,n);
> octave> d=ceil(logspace(2.3,3.5,n));
> octave> for i=1:n, s=rand(d(i)); tic; s'; t(i) = toc; end; # 
> loglog(d,t,'-x;;'); wpolyfit(log10(d.^2),log10(t),1)
> octave> wpolyfit(log10(d.^2),log10(t),1)
> Polynomial: 0.8061*x^1 - 5.31  [ p(chi^2>observed)=100.00% ]
> 
> But I had no time to interpret the results.
> 
> >Our cpu's are about the same age and specifications, and then,
> >this factor is probably not very important in the context of the
> >transposing operation. Remains the speed and amount of RAM, to
> >account for the wide difference between my timing and yours.
> 
> I have an Intel Pentium II 350 MHz, L@ cache 512 KB, 
> Asus P2B main board, Intel 440BX
> Dimm 512 MB pC 133 SDRAM certificate 133 MHz
> 
> the RAM has been acquired later, I think the board drives it at 100 MHz
> anyway, but maybe I twiddled with the parameters in the BIOS to reduce
> the access latency.  I can not be sure now, because I would need to
> switch off the PC, which I don't do easily.
> 
> -- 
> Francesco Potortì (ricercatore)        Voice: +39 050 315 3058 (op.2111)
> ISTI - Area della ricerca CNR          Fax:   +39 050 313 8091
> via G. Moruzzi 1, I-56124 Pisa         Email: address@hidden
> Web: http://fly.isti.cnr.it/           Key:   fly.isti.cnr.it/public.key

Hi,
Thanks for the answer. I have also used the equations given by P.
Kienzle, but I played with the logspace parameters to cover all the
range of interest with the desired points spacing. My last setting 
was given by:
t=zeros(1,n);d=ceil(logspace(2.5,3.6,n))
I also preferred to plot with plot(d,t./d.^2,'x') and changed
scales from linear to logarithmic using gnuplot's facilities.

As my swap partition is only 256Mb, and the memory requirements
of octave are roughly proportional with the square of d, octave
crached at values only slightly larger than 3.6. Your case may be
different.
Cheers, Avraham



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------