On 10/14/06, Andreas Schwarz <[EMAIL PROTECTED]> wrote:
> David Balmain wrote:
> > On 10/14/06, Andreas Schwarz <[EMAIL PROTECTED]> wrote:
> >> Ferret 0.10.11 & AAF: the time seems to be stored in a format that can't
> >> be sorted, the order doesn't make any sense. Workaround: use to_i on the
> >> Time object before putting it into the index.
> >
> > This is fine for sorting but it won't work when you want to do range
> > queries. If you want to run range queries against the field you will
> > need to pad the integer to a fixed width.
>
> Thanks, I didn't think of that. The width of a unix timestamp isn't
> going to change for the next few years, though.
>
> > I usually add dates in YYYYMMDD format so that they are also human
> > readable.
>
> I need at least seconds in this case.
>
> > Another thing to take into account is the precision you
> > want. The higher the precision you store the longer indexing and
> > searching will take. For small indexes however the difference will be
> > negligible.
>
> Will this also affect :sort performance? I have another index with 450k
> documents that I also would like to sort by time, that's probably no
> longer a small index. Is the sorting capable of :sorting and :limiting a
> large numer of results (e.g. 100k), or is this something that should
> only be done with small result sets?
It should sort the first time in under one second. After that the sort
will be cached so it will be even quicker. To use sort caching though,
be sure to use a sort object and not a sort_field or plain old string.
Also, for best performance, sort by byte instead of integer (which
will only work if you have a fixed width integer field). ie
:sort => Sort.new(SortField.new(:field_name, :type => :byte))
Cheers,
Dave
_______________________________________________
Ferret-talk mailing list
[email protected]
http://rubyforge.org/mailman/listinfo/ferret-talk
