From: Joaquin Cuenca Abela (e98cuenc@free.fr)
Date: Tue Jan 07 2003 - 15:32:35 EST
> -----Message d'origine-----
> De : owner-abiword-dev@abisource.com
> [mailto:owner-abiword-dev@abisource.com] De la part de Jesper Skov
> Envoyé : mardi 7 janvier 2003 19:44
> À : Joaquin Cuenca Abela
> Cc : AbiWord Dev List
> Objet : Re: RE : Piece Table with O(log(n)) operations
>
>
> On Mon, 2003-01-06 at 23:24, Joaquin Cuenca Abela wrote:
> > Just a follow-up. I've put in the web page some profile data (just
> > for the insert and delete operation, not yet anything complex).
> >
> > It sports a ~10 microseconds insertion and delete time for
> documents
> > of roughly 30,000 pages.
>
> What is the time for the old implementation?
It was too slow to measure it, so I measured a little piece table, and
then extrapolated the line (it was quite a perfect line).
It's ~0.6 seconds for the insertion (that's 60,000 times slower). As I
put in the web page, my double linked list implementation is a simpler
one than the AbiWord implementation... but for *this* test, my
implementation is faster than the real abiword implementation (because I
don't create a big vector by insertion).
I didn't measured the delete times so exactly as the insert operation,
but my little test extrapolates to ~0.5 seconds
(for a delete on a piece table with 250,000 pieces). So it's also
60,000 times slower.
As I said, I *don't* consider the piece table as our current bottleneck,
and the numbers prove my intuition to be right (remember that the test
if for a insanely huge piece table).
Cheers,
This archive was generated by hypermail 2.1.4 : Tue Jan 07 2003 - 15:36:43 EST