From: Dmitrey <no_valid@no.no>
Path: news.mathworks.com!newsfeed-00.mathworks.com!webcrossing
Newsgroups: comp.soft-sys.matlab
Subject: Re:
Message-ID: <ef455af.10@webcrossing.raydaftYaTP>
Date: Wed, 29 Nov 2006 10:14:23 -0500
References: <ef455af.5@webcrossing.raydaftYaTP> <ef455af.6@webcrossing.raydaftYaTP> <ef455af.7@webcrossing.raydaftYaTP> <ef455af.8@webcrossing.raydaftYaTP> <ef455af.9@webcrossing.raydaftYaTP>
Lines: 35
NNTP-Posting-Host: 193.239.179.130
MIME-Version: 1.0
Content-Type: text/plain; charset="ISO-8859-1"
Content-Transfer-Encoding: 8bit
Xref: news.mathworks.com comp.soft-sys.matlab:381046



Hi Ruslan,
of course I had tried with 2 workers too.
your second note is obvious & don't need further explain
the problem is that MATLAB distribution technic greatly fails the
comparison to, for example, C++ vfork().
As for me (and many other people) - I deal with problems with small
time*Big number of executions (numerical optimization, where
Objective function (along with constraints in other functions) is
called thouzands times). So using dfeval only greately slows the
process (usually objfun is calculated in dozens or hundreds times
faster than MATLAB parallel mechanizm, even for some functions with
nvars~1000).
no_valid@ukr.net precizely means the same - "no valid" :)
you can use my icq 275976670
WBR, D

 Ruslan wrote:
> at first: if you have dual core, you have to start 2 workers on
> your
> processor! not only one!
>
> at second: the examle you have explainde is not demonstrativ,
> because
> much more times need to create a task than computation itself! You
> will see performance if your calculation time is more than 60-80
> sec.
> There are some demos in MATLAB (portfolio optimization for
> example),
> it takes 80 seconds on one core, and 40 sec if 2 core are used.
>
> at third: your email no_valid@ukr.net is not valid.
>
> Hope it helps!
>
> feel free to contact me on ruslanp@softline.ru