[
https://issues.apache.org/jira/browse/LUCENE-1536?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Robert Muir updated LUCENE-1536:
--------------------------------
Attachment: LUCENE-1536_hack.patch
hack patch that computes the heuristic up front in weight init, so it scores
all segments consistently and returns the proper scoresDocsOutOfOrder for BS1.
Uwe's new test (the nestedFilterQuery) doesnt pass yet, don't know why.
I recomputed the benchmarks:
{noformat}
Task QPS trunkStdDev trunk QPS patchStdDev patch Pct
diff
PhraseF1.0 11.99 0.20 7.79 0.23 -37% -
-31%
TermF0.5 135.14 7.62 116.57 0.36 -18% -
-8%
AndHighHighF100.0 17.34 0.78 15.44 0.15 -15% -
-5%
AndHighHighF95.0 17.28 0.66 15.48 0.17 -14% -
-5%
AndHighHighF90.0 17.31 0.76 15.58 0.19 -14% -
-4%
AndHighHighF99.0 17.05 1.02 15.45 0.17 -15% -
-2%
AndHighHighF75.0 17.47 0.78 16.03 0.15 -12% -
-3%
AndHighHighF5.0 20.69 0.95 19.78 0.23 -9% -
1%
AndHighHighF1.0 35.11 1.46 33.64 0.36 -8% -
1%
AndHighHighF0.1 136.04 3.70 132.00 1.41 -6% -
0%
AndHighHigh 18.25 0.70 17.74 0.20 -7% -
2%
AndHighHighF0.5 49.84 1.72 48.58 0.49 -6% -
1%
TermF0.1 351.18 11.01 345.85 1.73 -4% -
2%
Fuzzy2F100.0 95.52 4.21 94.33 2.07 -7% -
5%
SloppyPhraseF100.0 8.01 0.28 7.91 0.09 -5% -
3%
Fuzzy2F90.0 95.42 3.86 94.51 1.74 -6% -
5%
Fuzzy2F95.0 95.20 4.86 94.33 1.83 -7% -
6%
Fuzzy1F1.0 54.02 1.67 53.56 1.07 -5% -
4%
PhraseF2.0 7.73 0.07 7.68 0.18 -3% -
2%
SloppyPhraseF99.0 7.99 0.23 7.95 0.10 -4% -
3%
AndHighHighF50.0 17.54 0.79 17.46 0.12 -5% -
4%
Fuzzy2F0.1 105.39 3.93 105.34 3.74 -7% -
7%
SpanNearF100.0 3.16 0.06 3.16 0.04 -2% -
2%
Fuzzy2F99.0 94.02 6.86 94.21 1.97 -8% -
10%
Fuzzy2F75.0 95.56 3.51 95.76 2.02 -5% -
6%
WildcardF2.0 52.79 0.27 53.05 0.57 -1% -
2%
Fuzzy1F0.5 58.12 1.83 58.43 1.22 -4% -
5%
PhraseF0.1 66.34 0.78 66.73 1.68 -3% -
4%
SloppyPhraseF0.1 56.15 1.52 56.79 0.64 -2% -
5%
SloppyPhrase 8.08 0.26 8.18 0.08 -2% -
5%
PKLookup 176.59 5.07 178.96 5.71 -4% -
7%
SpanNearF0.1 32.36 0.56 32.83 0.54 -1% -
4%
OrHighHighF0.1 78.20 0.52 79.44 0.74 0% -
3%
SloppyPhraseF95.0 7.91 0.08 8.05 0.09 0% -
3%
Fuzzy2 94.87 3.72 96.49 1.62 -3% -
7%
OrHighHighF0.5 31.41 0.47 31.96 0.33 0% -
4%
SpanNearF99.0 3.12 0.06 3.18 0.03 0% -
4%
WildcardF0.5 61.97 0.56 63.28 0.82 0% -
4%
PhraseF0.5 19.78 0.26 20.29 0.31 0% -
5%
SpanNear 3.19 0.08 3.27 0.05 -1% -
6%
WildcardF0.1 67.45 0.64 69.24 0.89 0% -
4%
SloppyPhraseF90.0 8.00 0.29 8.21 0.12 -2% -
8%
SpanNearF95.0 3.13 0.04 3.23 0.03 1% -
5%
Wildcard 43.19 0.34 44.64 1.40 0% -
7%
Fuzzy2F50.0 95.12 4.22 98.69 2.28 -2% -
11%
Fuzzy1 55.28 4.53 57.68 0.76 -4% -
15%
OrHighHigh 12.13 0.99 12.71 0.43 -6% -
18%
Phrase 3.60 0.04 3.81 0.04 3% -
7%
SpanNearF90.0 3.15 0.05 3.35 0.04 3% -
9%
Term 71.69 0.40 76.53 4.13 0% -
13%
PhraseF99.0 3.43 0.03 3.68 0.04 5% -
9%
PhraseF100.0 3.39 0.05 3.67 0.04 5% -
10%
SloppyPhraseF75.0 8.04 0.26 8.74 0.12 3% -
13%
Fuzzy2F20.0 97.38 4.03 106.17 2.88 1% -
16%
PhraseF95.0 3.38 0.03 3.70 0.05 6% -
11%
PhraseF90.0 3.42 0.02 3.76 0.03 8% -
11%
Fuzzy2F10.0 97.19 3.69 109.27 3.23 5% -
20%
PhraseF75.0 3.44 0.02 3.94 0.04 12% -
16%
Fuzzy2F5.0 96.77 4.17 112.60 3.30 8% -
25%
Fuzzy1F0.1 73.61 2.43 86.22 2.79 9% -
25%
WildcardF100.0 35.49 0.33 41.92 1.05 14% -
22%
SpanNearF75.0 3.15 0.07 3.72 0.04 14% -
22%
WildcardF95.0 35.43 0.24 41.90 0.99 14% -
21%
WildcardF90.0 35.59 0.32 42.11 1.11 14% -
22%
WildcardF99.0 35.43 0.34 41.94 1.09 14% -
22%
Fuzzy1F99.0 47.41 1.79 56.45 0.78 13% -
25%
WildcardF75.0 35.64 0.29 42.51 0.87 15% -
22%
Fuzzy1F100.0 46.85 1.83 56.42 0.55 14% -
26%
Fuzzy2F1.0 96.75 3.75 116.85 4.32 11% -
30%
Fuzzy1F95.0 46.91 1.37 56.69 0.69 15% -
25%
Fuzzy2F0.5 97.33 4.15 117.64 4.17 11% -
30%
Fuzzy2F2.0 95.51 3.65 115.66 3.95 12% -
30%
Fuzzy1F90.0 46.84 1.95 56.83 0.78 14% -
28%
WildcardF50.0 36.28 0.23 44.23 0.58 19% -
24%
TermF1.0 93.99 4.90 114.60 0.49 15% -
29%
Fuzzy1F75.0 47.12 1.68 58.11 0.82 17% -
29%
WildcardF1.0 56.94 0.80 71.15 0.80 21% -
28%
PhraseF50.0 3.49 0.01 4.39 0.04 24% -
27%
SloppyPhraseF50.0 8.03 0.28 10.12 0.13 20% -
32%
Fuzzy1F50.0 46.64 2.22 60.78 0.95 22% -
38%
OrHighHighF1.0 24.15 0.35 32.46 0.25 31% -
37%
WildcardF20.0 40.55 0.30 55.72 0.73 34% -
40%
Fuzzy1F20.0 49.29 1.52 69.91 1.26 35% -
48%
WildcardF5.0 47.02 0.33 67.11 0.81 40% -
45%
PhraseF5.0 5.03 0.06 7.29 0.13 40% -
49%
WildcardF10.0 43.04 0.57 62.68 0.69 42% -
49%
SpanNearF50.0 3.16 0.07 4.77 0.06 45% -
56%
AndHighHighF20.0 17.76 0.64 27.44 0.25 47% -
61%
Fuzzy1F10.0 48.53 2.47 75.31 1.60 44% -
66%
Fuzzy1F5.0 50.45 1.70 79.01 2.08 47% -
66%
Fuzzy1F2.0 52.03 1.54 82.41 2.46 49% -
68%
OrHighHighF100.0 7.69 0.20 12.19 0.33 50% -
67%
OrHighHighF99.0 7.72 0.35 12.25 0.34 47% -
70%
PhraseF20.0 3.74 0.03 5.95 0.05 56% -
61%
OrHighHighF95.0 7.79 0.28 12.43 0.33 49% -
69%
OrHighHighF2.0 19.60 0.24 31.28 0.14 56% -
62%
TermF2.0 70.16 3.78 112.18 0.53 51% -
69%
OrHighHighF90.0 7.83 0.23 12.56 0.33 51% -
69%
PhraseF10.0 4.14 0.04 6.76 0.09 59% -
66%
TermF50.0 42.57 1.77 70.63 1.25 56% -
76%
TermF75.0 41.43 1.61 70.40 2.44 57% -
82%
OrHighHighF75.0 7.81 0.22 13.33 0.36 61% -
80%
TermF95.0 41.07 1.65 70.96 3.30 58% -
88%
TermF99.0 41.12 1.57 71.14 3.41 58% -
88%
TermF90.0 40.92 1.58 70.81 3.00 59% -
87%
TermF100.0 40.01 0.73 71.10 3.36 66% -
89%
OrHighHighF50.0 8.39 0.24 14.92 0.29 69% -
86%
AndHighHighF10.0 18.68 0.61 36.17 0.23 86% -
101%
TermF20.0 45.66 1.98 88.55 0.52 84% -
103%
OrHighHighF5.0 14.89 0.30 29.12 0.20 90% -
100%
SloppyPhraseF20.0 8.34 0.29 16.36 0.26 86% -
106%
TermF5.0 52.71 1.88 105.42 0.46 92% -
108%
TermF10.0 47.53 1.99 97.62 0.51 96% -
115%
AndHighHighF2.0 26.32 1.16 54.83 0.27 98% -
119%
OrHighHighF20.0 10.36 0.19 22.12 0.22 107% -
119%
OrHighHighF10.0 12.31 0.35 26.43 0.23 106% -
122%
SloppyPhraseF10.0 8.73 0.28 22.70 0.38 147% -
172%
SloppyPhraseF0.5 19.21 0.58 52.13 0.61 160% -
183%
SpanNearF20.0 3.20 0.05 8.98 0.16 171% -
189%
SloppyPhraseF5.0 9.30 0.33 30.17 0.44 208% -
241%
SloppyPhraseF1.0 13.84 0.44 46.77 0.64 223% -
253%
SloppyPhraseF2.0 11.00 0.36 39.85 0.54 246% -
279%
SpanNearF10.0 3.31 0.07 13.57 0.23 294% -
325%
SpanNearF0.5 9.25 0.14 39.86 0.39 320% -
341%
SpanNearF5.0 3.54 0.07 19.35 0.38 425% -
468%
SpanNearF1.0 6.15 0.11 34.27 0.48 439% -
474%
SpanNearF2.0 4.52 0.09 28.11 0.43 500% -
543%
{noformat}
> if a filter can support random access API, we should use it
> -----------------------------------------------------------
>
> Key: LUCENE-1536
> URL: https://issues.apache.org/jira/browse/LUCENE-1536
> Project: Lucene - Java
> Issue Type: Improvement
> Components: core/search
> Affects Versions: 2.4
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Priority: Minor
> Labels: gsoc2011, lucene-gsoc-11, mentor
> Fix For: 4.0
>
> Attachments: CachedFilterIndexReader.java, LUCENE-1536-rewrite.patch,
> LUCENE-1536-rewrite.patch, LUCENE-1536-rewrite.patch,
> LUCENE-1536-rewrite.patch, LUCENE-1536-rewrite.patch,
> LUCENE-1536-rewrite.patch, LUCENE-1536-rewrite.patch,
> LUCENE-1536-rewrite.patch, LUCENE-1536.patch, LUCENE-1536.patch,
> LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch,
> LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch,
> LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch,
> LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch,
> LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch,
> LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch, LUCENE-1536.patch,
> LUCENE-1536_hack.patch, changes-yonik-uwe.patch, luceneutil.patch
>
>
> I ran some performance tests, comparing applying a filter via
> random-access API instead of current trunk's iterator API.
> This was inspired by LUCENE-1476, where we realized deletions should
> really be implemented just like a filter, but then in testing found
> that switching deletions to iterator was a very sizable performance
> hit.
> Some notes on the test:
> * Index is first 2M docs of Wikipedia. Test machine is Mac OS X
> 10.5.6, quad core Intel CPU, 6 GB RAM, java 1.6.0_07-b06-153.
> * I test across multiple queries. 1-X means an OR query, eg 1-4
> means 1 OR 2 OR 3 OR 4, whereas +1-4 is an AND query, ie 1 AND 2
> AND 3 AND 4. "u s" means "united states" (phrase search).
> * I test with multiple filter densities (0, 1, 2, 5, 10, 25, 75, 90,
> 95, 98, 99, 99.99999 (filter is non-null but all bits are set),
> 100 (filter=null, control)).
> * Method high means I use random-access filter API in
> IndexSearcher's main loop. Method low means I use random-access
> filter API down in SegmentTermDocs (just like deleted docs
> today).
> * Baseline (QPS) is current trunk, where filter is applied as iterator up
> "high" (ie in IndexSearcher's search loop).
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
