[ 
https://issues.apache.org/jira/browse/LUCENE-6015?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14177418#comment-14177418
 ] 

Adrien Grand commented on LUCENE-6015:
--------------------------------------

Here are the results of luceneutil benchmarks for various values of the shift 
(baseline is trunk with a shift of 14, the patch version has a modified shift).

Shift = 12
{noformat}
                Wildcard       18.07     (15.9%)       17.71     (14.6%)   
-2.0% ( -28% -   34%)
                  IntNRQ        7.02     (16.5%)        6.89     (15.1%)   
-1.9% ( -28% -   35%)
                 Prefix3      186.96     (12.6%)      184.80     (12.1%)   
-1.2% ( -22% -   26%)
                  Fuzzy1       81.78     (14.3%)       81.52     (12.9%)   
-0.3% ( -24% -   31%)
                  Fuzzy2       45.23     (13.5%)       45.33     (13.6%)    
0.2% ( -23% -   31%)
{noformat}

Shift = 10
{noformat}
                Wildcard      117.85     (11.6%)      116.78     (10.4%)   
-0.9% ( -20% -   23%)
                 Prefix3       36.49     (13.7%)       36.89     (13.6%)    
1.1% ( -23% -   32%)
                  IntNRQ       21.47     (13.5%)       21.91     (13.6%)    
2.0% ( -22% -   33%)
                  Fuzzy1       91.55     (12.6%)       96.88     (10.7%)    
5.8% ( -15% -   33%)
                  Fuzzy2       50.22     (15.6%)       53.28     (12.0%)    
6.1% ( -18% -   39%)
{noformat}

Shift = 9
{noformat}
                  Fuzzy1       81.66      (5.5%)       81.43      (8.0%)   
-0.3% ( -13% -   13%)
                 Prefix3       25.41      (4.8%)       25.75      (4.0%)    
1.4% (  -7% -   10%)
                Wildcard       20.52      (4.2%)       20.88      (3.1%)    
1.8% (  -5% -    9%)
                  IntNRQ       10.63      (4.9%)       10.98      (4.3%)    
3.3% (  -5% -   13%)
                  Fuzzy2       61.61     (10.7%)       63.89     (11.8%)    
3.7% ( -16% -   29%)
{noformat}

Shift = 8
{noformat}
                  IntNRQ        9.30     (12.8%)        9.01     (11.5%)   
-3.0% ( -24% -   24%)
                Wildcard       99.80      (9.7%)       96.97      (8.5%)   
-2.8% ( -19% -   17%)
                 Prefix3       80.28     (11.1%)       78.95     (10.0%)   
-1.7% ( -20% -   21%)
                  Fuzzy1       86.77     (10.9%)       88.09      (9.3%)    
1.5% ( -16% -   24%)
                  Fuzzy2       70.89      (9.7%)       73.07     (10.9%)    
3.1% ( -15% -   26%)
{noformat}

Shift = 7
{noformat}
                 Prefix3      128.85     (15.8%)      103.20      (1.8%)  
-19.9% ( -32% -   -2%)
                  IntNRQ       10.91     (17.4%)       11.12      (3.7%)    
1.9% ( -16% -   27%)
                Wildcard       31.95     (16.8%)       34.15      (2.9%)    
6.9% ( -10% -   31%)
                  Fuzzy2       51.60     (14.9%)       55.55      (3.7%)    
7.6% (  -9% -   30%)
                  Fuzzy1       73.90     (14.5%)       79.94      (2.9%)    
8.2% (  -8% -   29%)
{noformat}

Shift = 6
{noformat}
                Wildcard       50.46      (9.1%)       38.86      (6.4%)  
-23.0% ( -35% -   -8%)
                 Prefix3       45.21      (9.6%)       38.48      (7.8%)  
-14.9% ( -29% -    2%)
                  IntNRQ        8.60     (10.3%)        7.53      (8.5%)  
-12.4% ( -28% -    7%)
                  Fuzzy1       96.87     (10.6%)       97.11      (9.9%)    
0.2% ( -18% -   23%)
                  Fuzzy2       64.50     (12.6%)       67.21     (12.4%)    
4.2% ( -18% -   33%)
{noformat}

> Revisit DocIdSetBuilder's heuristic to switch to FixedBitSet
> ------------------------------------------------------------
>
>                 Key: LUCENE-6015
>                 URL: https://issues.apache.org/jira/browse/LUCENE-6015
>             Project: Lucene - Core
>          Issue Type: Bug
>            Reporter: Adrien Grand
>            Assignee: Adrien Grand
>            Priority: Minor
>
> DocIdSetBuilder starts with a SparseFixedBitSet and then upgrades to a 
> FixedBitSet when the cardinality grows larger than maxDoc >>> 14. However 
> Robert improved SparseFixedBitSet performance quite significantly in 
> LUCENE-6003 so we should see if it makes sense to update this heuristic.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to