[ZODB-Dev] Another possible ZODB optimization

[Toby Dickenson]

On Thursday 09 January 2003 8:53 pm, Greg Ward wrote:

> Since Jim and
> Jeremy have already gone to some effort to do so for ZODB 3.2 --
> specifically, by replacing the in-memory object index with a BTree --

> Anyways, initial results look promising: I wrote a script that creates
> either a FastIndex object (my new type) or a plain dictionary, and
> adds consecutive integers to it, eg.
>
>   index['\0\0\0\0\0\0\0\0'] =3D ...
>   index['\0\0\0\0\0\0\0\1'] =3D ...
>
> Off the cuff, it looks like FastIndex is about 10% faster and 3x smalle=
r
> than a regular dictionary.  Specifically, to add 750,000 entries:
>
>   dict        4.5 sec     64804 kB RSS
>   FastIndex   4.1 sec     18568 kB RSS

How does Jim's fsIndex compare at this?

--=20
Toby Dickenson
http://www.geminidataloggers.com/people/tdickenson