incubator-ooo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Rob Weir <robw...@apache.org>
Subject Re: Performance!
Date Thu, 10 May 2012 15:31:26 GMT
On Thu, May 10, 2012 at 11:06 AM, drew <drew@baseanswers.com> wrote:
> On Thu, 2012-05-10 at 08:51 -0400, Rob Weir wrote:
>> On Thu, May 10, 2012 at 8:10 AM, Rob Weir <robweir@apache.org> wrote:
>> > On Thu, May 10, 2012 at 6:53 AM, Ross Gardler
>> > <rgardler@opendirective.com> wrote:
>> >> Thanks Imacat,
>> >>
>> >> This was originally posted to the private list so as not to offend
>> >> some of our more sensitive list subscribers. However, some useful
>> >> discussion started looking at why the graphs looked like they did. I,
>> >> as a mentor, requested that it be moved here so that everyone, could
>> >> benefit from the discussion. Imacat did not post all comments, only
>> >> the link that was the catalyst, since they were made in private, it's
>> >> up to others to bring their constructive thoughts here.
>> >>
>> >> I think I see a potential for collaboration between the various ODF
>> >> related projects here.
>> >>
>> >> Can a few sample documents be created which produce graphs showing
>> >> better performance in other ODF products? Michael, you say they can do
>> >> that for LO, I invite you to do so. Such documents would help AOO
>> >> developers explore weakness in AOO code.
>> >>
>> >> At the same time AOO could provide documents that demonstrate better
>> >> AOO performance. These will help other projects explore weaknesses in
>> >> their own  code.
>> >>
>> >> RANDOM THOUGHT: are there any ODF test documents that might serve this purpose?
>> >>
>> >
>> > Another idea:  the blog post also indicates that AOO 3.4 uses less RAM
>> > than LO:  35Mb versus 43MB.   This might be related to the start up
>> > performance difference.  But since neither product has made radical
>> > changes to internal memory structures, any difference in memory
>> > consumption is probably related to what libraries are loaded at
>> > startup.  That should be easier to track down.
>> >
>> > Also, a comparison of AOO 3.4 versus OOo 3.3.0 would indicate whether
>> > we're dealing with a coding improvement in AOO 3.4 or a regression in
>> > LO.  Whatever the result,  that gives useful information that can be
>> > used to improve performance.
>> >
>>
>> A quick test suggests a little of both:
>>
>> Looking soffice.bin ("working set" memory footprint in Windows XP) for
>> Writer start up, no document loaded:
>>
>> OOo 3.3.0 = 95,792 Kb
>> AOO 3.4.0 = 88,508 Kb
>> LO  3.5.1 = 108,120 Kb
>>
>> So compared to OOo 3.3.0, AOO 3.4 is reduced 8% and LO increased 13%.
>>  Of course, RAM is (relatively) cheap, so the raw numbers are not that
>> important.  But any associated initialization code associated with
>> whatever is causing this difference, that could easily impact start
>> performance.
>>
>
> Alright - likely I don't need to ask this - The packages ship from a
> really different mindset, one Aoo is bare bones (particularly this
> specific release) and LibO comes with condiments.
>
> So - just to be sure, did you pull out the extras (the extensions) that
> come default with LibO, before checking the footprint?
>

Just the default install.  I didn't change anything.  And likely
neither did the user who was doing the timings.

If that explains the start up differences and the working set
differences, then that is good to know.   As engineers I think we all
struggle to find ways to add features without slowing the product
down.  We'll face the same set of issues.as we enhance AOO.

-Rob


> Thanks,
>
> //drew
>

Mime
View raw message