[jira] Updated: (LUCENE-1593) Optimizations to TopScoreDocCollector and TopFieldCollector

["Shai Erera (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/LUCENE-1593?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Shai Erera updated LUCENE-1593:
-------------------------------

    Attachment: PerfTest.java
                LUCENE-1593.patch

The patch implements all that has been suggested except:
* pre-populating the queue in TopFieldCollector - as was noted here previously, this seems to remove the 'if (queueFull)' check but add another 'if' in FieldComparator (which may be executed several times per collect().
* Move initCountingSumScorer() to BS2's ctor and add(). That's because if more than one Scorer is added we create a DisjunctionSumScorer, which initializes its queue by calling next() on the passed-in Scorer. Therefore if we call initCountingSumScorer for every Scorer added, we advance that Scorer as well as all the previous ones. I chose to discard that optimization, which only affects next() and skipTo().

The patch also includes the fix for TestSort in the 2.4 back_compat branch. I only fixed TestSort, and not MultiSearcher and ParallelMultiSearcher.

All tests pass.

I also ran some performance measurements (all on SRV 2003):

|| JRE || sort || best time (trunk) || best time (patch) || diff (%) ||
| SUN 1.6 | int | 1017.59 | 1015.96 | {color:green}~1%{color} |
| SUN 1.6 | doc | 767.49 | 763.20 | {color:green}~1%{color} |
| IBM 1.5 | int | 1018.77 | 1017.39 | {color:green}~1%{color} |
| IBM 1.5 | doc | 768.10 | 764.14 | {color:green}~1%{color} |

As you can see, there is a slight performance improvement, but nothing too dramatic.

You are welcome to review the patch as well as run the PerfTest I attached. It accepts two arguments: <indexDir> and [sort]. 'sort' is optional and if not defined it sorts by doc. Otherwise, whatever you pass there, it sorts by int.

> Optimizations to TopScoreDocCollector and TopFieldCollector
> -----------------------------------------------------------
>
>                 Key: LUCENE-1593
>                 URL: https://issues.apache.org/jira/browse/LUCENE-1593
>             Project: Lucene - Java
>          Issue Type: Improvement
>          Components: Search
>            Reporter: Shai Erera
>             Fix For: 2.9
>
>         Attachments: LUCENE-1593.patch, PerfTest.java
>
>
> This is a spin-off of LUCENE-1575 and proposes to optimize TSDC and TFC code to remove unnecessary checks. The plan is:
> # Ensure that IndexSearcher returns segements in increasing doc Id order, instead of numDocs().
> # Change TSDC and TFC's code to not use the doc id as a tie breaker. New docs will always have larger ids and therefore cannot compete.
> # Pre-populate HitQueue with sentinel values in TSDC (score = Float.NEG_INF) and remove the check if reusableSD == null.
> # Also move to use "changing top" and then call adjustTop(), in case we update the queue.
> # some methods in Sort explicitly add SortField.FIELD_DOC as a "tie breaker" for the last SortField. But, doing so should not be necessary (since we already break ties by docID), and is in fact less efficient (once the above optimization is in).
> # Investigate PQ - can we deprecate insert() and have only insertWithOverflow()? Add a addDummyObjects method which will populate the queue without "arranging" it, just store the objects in the array (this can be used to pre-populate sentinel values)?
> I will post a patch as well as some perf measurements as soon as I have them.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


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