Re: Poor plot performance on Windows

[Nicholas Jankowski]

On Wed, Mar 16, 2016 at 9:08 AM, Tatsuro MATSUOKA <address@hidden> wrote:

----- Original Message -----

> From: rcharan51 
> To: help-octave
> Cc:
> Date: 2016/3/16, Wed 15:49
> Subject: Re: Poor plot performance on Windows

>
> I am using octave 4 on win 10. Every time octave crashes while printing a
> 3d surface be it with or without legends. So not everyone is content with
> plot performance please help with some work around.
> On Mar 16, 2016 12:42, "Jens.Wulf [via Octave]" <
> address@hidden> wrote:
>>
>>  Hi Jake,
>>
>>  I did some tests in that area. Results:
>>   - versions after 3.2.3 (windows) or 3.2.4 (linux) are increasingly slow
>>   - win is slower than linux
>>   - with versions after 3.2.x, it makes a difference on how you adjust
> your legend. Please see my
>>    posts from some days ago ('Slow plotting (redraw due to
> legend?)') for
> details. Adjusting defaults in legend.m
>>    might also save you about a third of your plotting time.
>>   - try not to use 'hold on', many plots, 'hold off'. Time
> increases
> linearily with the number of plots.
>>   - I tested gnuplot, fltk and qt with different versions of octave and
> many different uses of plot(), legend(),
>>    linetypes and so on.  As far as speed is concerned, they were equally
> slow. The only difference I remember
>>    was when using drawnow(), which made gnuplot slower than the others.
>>
>>  My posts caused little echo on this list; it seems that most people are
> content with plot performance. For my interactive plotting usecase,
> everything after 3.2.x is unusable.
>>
>>  Best regards,
>>  Jens
>>
>>
>>  ________________________________
>>
>>
>>  SEW-EURODRIVE GmbH & Co KG
>>  Kommanditgesellschaft, Sitz: Bruchsal, RG Mannheim HRA 230970
>>  Komplementärin: SEW-EURODRIVE Verwaltungs-GmbH, Sitz: Bruchsal, RG
> Mannheim HRB 230207
>>
>>  Gesellschafter: Jürgen Blickle, Rainer Blickle
>>  Geschäftsführer: Jürgen Blickle (Vorsitzender), Udo Aull, Johann Soder,
> Dr. Jürgen Zanghellini
>>  _______________________________________________
>>  Help-octave mailing list
>>  [hidden email]
>>  https://lists.gnu.org/mailman/listinfo/help-octave
>>
>>
>>  ________________________________
>>  If you reply to this email, your message will be added to the discussion
> below:
>>
> http://octave.1599824.n4.nabble.com/Poor-plot-performance-on-Windows-tp4675531p4675532.html
>>  To start a new topic under Octave - General, email
> address@hidden
>>  To unsubscribe from Octave, click here.
>>  NAML
>

Slowness of plotting on windows has been frequently asked.
Note that not all people claim the slowness.

Most people have experience of the slowness only on the first plot.
Perhaps this comes from fontconfig cash problem.
The similar cases happen on gnuplot for windows in case of cairo and gd based
terminals in which fontconfig system is used.

Octave graphics toolkit also uses the fontconfig system then I suspect that
slowness of the first plot comes from the fontconfig system.

However, there are not small number of people who claim the incredible 
slowness on octave for windows.

Difficulty of the issue is that not all people experience of the incredible slowness.
I have never  experienced the slowness issue except for the first plot on
several PCs at home and office.

To overcome the issue development team has to know where is the bottle neck.t
That is very difficult.

One possibility to avoid of slowness of the plot is use octave on Cygwin.
I sometimes uses octave on Cygwin.
It is well prepared with extensive efforts by Marco Atzeri and feel it not bad.

Please see the instruction for Octave 4.0.0 on cygwin

in

http://wiki.octave.org/Octave_for_Microsoft_Windows


Regards

Tatsuro

I think Tatsuro summarizes the problem pretty well. The majority of developers are not running Windows, and do not have a way to reliably recreate and troubleshoot the problem. Some of those on Windows who see the problem don't see it repeatably. Some of those who do see it are not developers, nor have they been able to produce a repeatable, simple case demonstrating the problem. (I have seen the problem. but it's intermittent. I saw it more under 3.8.2, and a few times under 4.0.0, but generally not at all now, and cannot explain why.)

If you are able to repeatably create the problem, maybe we can start paring down the problem description for those who might be able to look under the hood. Tatsuro mentioned one hypothesis about fontcache issues. Does anyone know how to isolate that in a sample script?

Jens's thread from a short while ago showed good examples in using the profiler to narrow down the problem, but I thought that one wound up only being partially due to the plotting delay.

I'm pretty sure the Octave 4.00. MXE build uses MINGW

nickj