TrieIntField vs IntPointField performance only for equality comparison (no range filtering)

[Dorian Hoxha <do...@gmail.com>]

Hi,

Has anyone measured which is more efficient/performant between the 2
intfields if we don't need to do range-checking ? (precisionStep=0)

Regards,
Dorian

Re: TrieIntField vs IntPointField performance only for equality comparison (no range filtering)

[Dorian Hoxha <do...@gmail.com>]

Hi Shawn,

I forgot that legacy-int-fields were deprecated. Point fields it is then.

Thanks,
Dorian

On Tue, May 16, 2017 at 3:01 PM, Shawn Heisey <ap...@elyograg.org> wrote:

> On 5/16/2017 3:33 AM, Dorian Hoxha wrote:
> > Has anyone measured which is more efficient/performant between the 2
> > intfields if we don't need to do range-checking ? (precisionStep=0)
>
> Point field support in Solr is *BRAND NEW*.  Very little information is
> available yet on the Solr implementation.  Benchmarks were done at the
> Lucene level, but I do not know what the numbers were.  If any Solr
> benchmarks were done, which I can't be sure about, I do not know where
> the results might be.
>
> Lucene had Points support long before Solr did.  The Lucene developers
> felt so strongly about the superiority of the Point implementations that
> they completely deprecated the legacy numeric field classes (which is
> what Trie classes use) early in the 6.x development cycle, slating them
> for removal in 7.0.
>
> If you wonder about backward compatibility in Solr 7.0 because the
> Lucene legacy numerics are disappearing, then you've discovered a
> dilemma that we're facing before the 7.0 release.
>
> Thanks,
> Shawn
>
>

Re: TrieIntField vs IntPointField performance only for equality comparison (no range filtering)

[Shawn Heisey <ap...@elyograg.org>]

On 5/16/2017 3:33 AM, Dorian Hoxha wrote:
> Has anyone measured which is more efficient/performant between the 2
> intfields if we don't need to do range-checking ? (precisionStep=0) 

Point field support in Solr is *BRAND NEW*.  Very little information is
available yet on the Solr implementation.  Benchmarks were done at the
Lucene level, but I do not know what the numbers were.  If any Solr
benchmarks were done, which I can't be sure about, I do not know where
the results might be.

Lucene had Points support long before Solr did.  The Lucene developers
felt so strongly about the superiority of the Point implementations that
they completely deprecated the legacy numeric field classes (which is
what Trie classes use) early in the 6.x development cycle, slating them
for removal in 7.0.

If you wonder about backward compatibility in Solr 7.0 because the
Lucene legacy numerics are disappearing, then you've discovered a
dilemma that we're facing before the 7.0 release.

Thanks,
Shawn