lucene-java-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Otis Gospodnetic <otis_gospodne...@yahoo.com>
Subject Re: Benchmarking LUCENE-584 with contrib/benchmark
Date Thu, 05 Apr 2007 01:35:36 GMT
Hi Doron,

Yes, this was great help, thanks!  I've got my:
1. MatchTask (just like ReadTask, but with searcher.match(Query, new MatchCollector() ....))
2. SearchMatchTask (just like SearchTask, but extends MatchTask), so I was able to use "SearchMatch"
in the alg file where "Search" was before.

I need to compare search with HitCollector to matching with MatchCollector, and leave both
of their their collect(...) methods empty in order to see how much time, if any, is shaved
off with Paul's patch that simply skips scoring.  Thus, I've modified ReadTask locally to
use the search method with HitCollector instead of the one that returns Hits.

I'll put both the new benchmark Tasks and my results in LUCENE-584 once I have something (tomorrow,
most likely).

Otis


----- Original Message ----
From: Doron Cohen <DORONC@il.ibm.com>
To: java-user@lucene.apache.org
Sent: Monday, April 2, 2007 8:20:58 PM
Subject: Re: Benchmarking LUCENE-584 with contrib/benchmark

Hi Otis, you could use the byTask package - add your-type-of-search-task.
Suffix the new task class name by "Task" - e.g. NewNameTask - and then you
can use the 'command' "NewName" in an alg file.

I am not sure you can extend/reuse the existing ReadTask for this, because
its implementation of search/retrieve/traverse assumes using certain
Lucene's API's.

Does this make sense?

Regards,
Doron

Otis Gospodnetic <otis_gospodnetic@yahoo.com> wrote on 02/04/2007 15:35:57:

> Hi,
>
> I'm looking at benchmarking Paul's http://issues.apache.
> org/jira/browse/LUCENE-584 code.
> I'd like to compare either:
>   HitCollector.collect(doc, score) vs. MatchCollector.collect(doc)
>   or
>   IndexSearcher.search(Weight, Filter, HitCollector) vs.
> IndexSearcher.match(Query, MatchCollector)
>
> ...and see what the performance difference is when one skips
> scoring.  I can cook up my own benchmark, but I'd rather try using
> contrib/benchmark, if it can be used for this purpose.  Can it?  I
> browsed around contrib/benchmark looking for a place where I could
> plug in calls to IndexSearcher.match(Query, MatchCollector), and
> found that the only search calls are in the  StandardBenchmarker
> class.  From what I can tell, I can't really plug in any other calls
> in there, can I?  It looks like I simply need to write my own
> MatcherBenchmarker and implement my own benchmark(File,
> BenchmarkOptions) method.  Once that's in place, I'd have to run one
> benchmark with the StandardBenchmarker and the other one with my
> MatcherBenchmarker and manually compare the numbers.
>
> Is that the right approach, or am I missing a shortcut?
>
> Thanks,
> Otis


---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org





---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe@lucene.apache.org
For additional commands, e-mail: java-user-help@lucene.apache.org


Mime
View raw message