Re: Very slow mesh on MacOS X

[Quentin Spencer]

John W. Eaton wrote:
> On  6-Oct-2006, Robert A. Macy wrote:
>
> | I use octave 2.1.50a on Windows 98
> | 
> | and *never* use mesh, way too slow
> | 
> | I write special scripts to display meshed items, using the
> | gsplot commands
> | 
> | display 200 by 1000 in around 1/2 second on a 750MHz
> | machine
> | 
> | I found mesh only is fast enough on 20 by 40, or such
>
> Since mesh eventually just invokes gnuplot, it seems odd to me that it
> would be too slow and your calls to gnuplot are acceptably fast, but
> mesh does have a for loop, so maybe that is the source of the problem
> for larger data sets.  Perhaps that can be eliminated.  But in any
> case, can you post an example that demonstrates the slow response
> time, and for which calls to gsplot to generate the same plot do not
> show the problem?
>
> It really doesn't help us to have a lot of posts claiming that Octave
> is slow without actual examples that can be used to try to diagnose
> and fix the problems.  Vague descriptions of the problems are no
> substitute for a precise example, since then all we can do is guess
> about what the problems might be.  Often the guesses are wrong and
> just a waste of time.
>   

I've noticed slowness with large mesh plots as well, but I'm inclined to 
think the problem is in gnuplot. It can be seen pretty easily using the 
sombrero function. On my machine sombrero(40) gives a nearly 
instantaneous result, sombrero(100) is a little slower, and 
sombrero(200) results in a considerably longer wait. I think it's 
gnuplot for two reasons: first all three commands return to the prompt 
fairly quickly (quicker for smaller matrices of course), and second, 
after the sombrero(200) plot is up, resizing the window to full screen 
and back results in a significant delay while gnuplot re-draws.

Having said that, if there are for loops in the mesh code, it would be 
nice to eliminate them. Also, gnuplot 4.2 will have a binary data 
transfer capability, and I think using it could help with any plots (2D 
or 3D) that have large amounts of data, although I'm still not sure 
that's the bottleneck here.

Quentin