Hi Paul,
please see in line...
On Fri, Oct 05, 2007 at 12:57:34PM +0200, Paul Sandoz wrote:
> Jakub Podlesak wrote:
> >Hi Frank,
> >
> >Very impressive results! Congratulations.
> >
>
> Yes. It looks very promising. I am going to send am email out talking
> about the whole URI resolving aspects and how we may proceed forward.
>
>
> >And it is interesting,
> >that for the medium set (50 uris) the result is always
> >better than for the small sets (20 uris).
> >
>
> Do you mean the other way around? The larger the bar the better the result.
no, i mean the following (differencies are more apparent for the linear driver):
Linear Driver:
time result
Small set (20 templates) right handed 12.279 0.407
Medium set (50 templates) right handed 7.782 0.643 (bigger set, better result)
Small set (20 templates) right slashed 12.369 0.404
Medium set (50 templates) right slashed 8.081 0.619 --"--
Trie Driver:
Small set (20 templates) right handed 0.047 105.875
Medium set (50 templates) right handed 0.039 129.814 --"--
Small set (20 templates) right slashed 0.037 133.742
Medium set (50 templates) right slashed 0.036 137.403 --"--
>
> It is tricky to compare results between different sets because each set
> uses different input data (although the size is the same) and different
> URI templates. Plus the linear algorithm could potentially terminate
> on average less than for other tests depending on the set of URIs.
Absolutely. The important thing is that Frank's algorithm is much better
than the linear one.
>
>
> >Did you try yet another different uri sets? And different uri set sizes
> >(200, 500)?
> >
>
> Frank gave me the source for the drivers and i modified to test 2 4 8 16
> 32 64 128 templates using the same set source source data (reduced in
> size but i increased the number of warmup and run iterations). I plotted
> this of a log log scatter graph (see attached).
Nice graph.
~Jakub
>
> As you can see there is variance between linear and Trie between sets of
> data but the clear trend is that:
>
> 1) Trie is always faster;
>
> 2) Linear degrades when the number of templates increases;
>
> 3) Trie displays good scalability as the number of templates increases; and
>
> 4) Trie can get an order of magnitude better than linear for small sets
> of templates e.g. 8 templates.
>
> Paul.
>
> >~Jakub
> >
> >On Thu, Oct 04, 2007 at 05:42:11PM -0500, Frank Martínez wrote:
> >>Hi all guys,
> >>
> >>I have made some benchmarkings with TrieUriPathResolver and
> >>LinearOrderedUriPathResolver.
> >>
> >>Look some results in the (attached) japex report.
> >>Please read the (attached) README file before.
> >>
> >>Regards,
> >>
> >>--
> >>Frank D. Martínez M.
> >>Asimov Technologies Ltda.
> >>Blog: http://www.ibstaff.net/fmartinez/
> >
> >
> >
> >>---------------------------------------------------------------------
> >>To unsubscribe, e-mail: dev-unsubscribe_at_jersey.dev.java.net
> >>For additional commands, e-mail: dev-help_at_jersey.dev.java.net
> >
> >---------------------------------------------------------------------
> >To unsubscribe, e-mail: dev-unsubscribe_at_jersey.dev.java.net
> >For additional commands, e-mail: dev-help_at_jersey.dev.java.net
> >
>
> --
> | ? + ? = To question
> ----------------\
> Paul Sandoz
> x38109
> +33-4-76188109
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe_at_jersey.dev.java.net
> For additional commands, e-mail: dev-help_at_jersey.dev.java.net