En/na Taco Hoekwater ha escrit:

Xan wrote:

...

Hi,

I have Intel Core Duo and when I run texexec the CPU resource in gnome is marked about 50%. Is it mean that texexec only use one processor? If it's, can you think to improve it for scalability?

Just a curious question.

Run two jobs at once ;-)

Seriously: no, typesetting (currently) is a single-process streaming task, and that is not easily fixed at all.

Best wishes, Taco Confirmed so :-) Yes, I understand it's not easy to "fix" it. I think it's a huge code task.

It could be useful for offering typesetting services. I'm thinking about having a computation service like alpha.wolfram but with ConTeXt. Like ConTeXt online but benefit from multiple processors.... Joking, perhaps you could fix it in Mark VII ;-) Regards, Xan.

> You cannot start processing page 500 before you know where page 499
> wil end. TeX is not really a kind of application where you would gain
> a lot by parallelization. And honestly, I don't remember seing many
> applications using both cores. (Plus: I'm happy if other applications
> continue to run smoothly instead of being blocked by TeX using 95% of
> both processors.)
Hm, not so sure. Think for example to a book with 3 chapters, and you
know at priori that there are not relations
among them (references etc).
You can
1) typeset each of them in a concurrent way,
2) recompose the final document to correct pagenumbers and build indeces

(this is a TeX concurrency )

>
>> It could be useful for offering typesetting services. I'm thinking about
>> having a computation service like alpha.wolfram but with ConTeXt. Like
>> ConTeXt online but benefit from multiple processors....
>
> If you would offer a service, you would get multiple requests at the
> same time anyway, so there's no real need for that in this case
It's another kind of concurrency, it's about servers and OS (apache -
Linux vs IIE - Windows server etc)

> You can try XeTeX if you want to put load on both processors. It does
> offer a parallel process as far as I heard, but then you'll probably
> want support for quad-core once you get a better computer :)
Again, another kind of concurrency.
it's about luatex or pdftext or xetex C programs


-- 
luigi

On Fri, Aug 28, 2009 at 9:15 AM, Taco Hoekwater wrote:

luigi scarso wrote:

...

...

You cannot start processing page 500 before you know where page 499 wil end.  []

...

Hm, not so sure. Think for example to a book with 3 chapters, and you know at priori that there are not relations

The root of the problem is that because TeX is a programming language as well as a typesetting engine, there is no way of knowing what the state of the engine will be after the next token has been read, never mind the next chapter.

I'm think at something like this %% here Global macros and data (better if readonly) %% \StartTask \chaper{..} : \StopTask \StartTask \chaper{..} : \StopTask \StartTask \chaper{..} : \StopTask ie **you** marked that contents between \StartTask and \StopTask are suitable for concurrency -- luigi