[texhax] Throughput

[Peter Davis]

On Fri, Mar 4, 2011 at 2:38 PM, Peter Davis <pfd at pfdstudio.com> wrote:

>
>
> On Thu, Mar 3, 2011 at 4:07 PM, Peter Davis <pfd at pfdstudio.com> wrote:
>
>>
>> On Thu, Mar 3, 2011 at 10:53 AM, Martin Schröder <martin at oneiros.de>wrote:
>>
>>> 2011/3/3 William Adams <will.adams at frycomm.com>:
>>> > Have you tried processing it w/o including the graphics?
>>>
>>> That would be my first guest. Try with the draft option...
>>>
>>
>> (NB: Just replying to texhax, to avoid cross-posting issues I seem to have
>> triggered.)
>>
>> I tried running 200 pages, with and without draft mode.  Specifically, I
>> put draft mode on the graphicx package:
>>
>> \includepackage[draft]{graphicx}
>>
>> I used the --time-statistics option on the xelatex command to get timings.
>>  The results are:
>>
>> Non-draft:
>> gross: 111634 ms, user mode: 17050 ms, kernel mode: 50778 ms, total: 67828
>> Draft:
>> gross: 77941 ms, user mode: 12589 ms, kernel mode: 27300 ms, total: 39889
>>
>
> Now using \newfontfamily *and* [draft]{graphicx} ...
>
> gross execution time: 45458 ms
> user mode: 8658 ms, kernel mode: 25006 ms, total: 33664
>
> A definite improvement.  Now to try etex and box registers for the external
> graphics.
>


Ok, I tried using \savebox, and got this:

gross execution time: 3996 ms
user mode: 1123 ms, kernel mode: 436 ms, total: 1559

An order of magnitude improvement, that brings this into line with what I
need.  I still have to work out a hashing scheme to come up with unique
names for the images.  This test used a crappy scheme that had many name
collisions, so the "real" results will not be this good, since there will be
more unique images.

Thanks!
-pd


-- 
----
The Tech Curmudgeon
http://www.techcurmudgeon.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://tug.org/pipermail/texhax/attachments/20110304/32c9f364/attachment-0001.html>