[
https://issues.apache.org/jira/browse/LUCENE-4702?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17024532#comment-17024532
]
Adrien Grand commented on LUCENE-4702:
--------------------------------------
OK I benchmarked with multi-segment indices this time to try to better
replicate nightly benchmarks. I opened a pull request at
https://github.com/apache/lucene-solr/pull/1216 that:
- removes compression of suffix lengths since it didn't help much anymay,
- replaces LZ4 on stats by explicit run-length compression
- only tries out LZ4 for suffix bytes if the average suffix length is > 6 to
reduce index-time overhead since it's unlikely to meet the saving expectations
otherwise anyway, in order to reduce index-time overhead
On wikibigall, the specialized RLE makes the tim file even smaller with this
change (969MB vs. 996MB) and luceneutil seems to be a bit more happy:
{noformat}
TaskQPS baseline StdDev QPS patch StdDev
Pct diff
IntNRQ 144.16 (1.2%) 143.47 (1.9%)
-0.5% ( -3% - 2%)
TermBGroup1M 32.04 (5.1%) 31.93 (5.1%)
-0.4% ( -10% - 10%)
TermDTSort 39.13 (0.9%) 39.05 (1.0%)
-0.2% ( -2% - 1%)
TermGroup1M 40.18 (4.0%) 40.12 (3.4%)
-0.2% ( -7% - 7%)
TermTitleSort 124.62 (1.9%) 124.54 (1.6%)
-0.1% ( -3% - 3%)
TermDayOfYearSort 88.37 (6.9%) 88.34 (7.1%)
-0.0% ( -13% - 14%)
TermGroup10K 28.56 (5.0%) 28.56 (4.4%)
0.0% ( -8% - 9%)
IntervalsOrdered 4.50 (1.1%) 4.51 (0.6%)
0.0% ( -1% - 1%)
TermBGroup1M1P 45.83 (4.1%) 45.85 (4.0%)
0.0% ( -7% - 8%)
TermMonthSort 137.33 (1.8%) 137.40 (1.3%)
0.1% ( -2% - 3%)
AndHighHigh 72.97 (2.8%) 73.05 (2.7%)
0.1% ( -5% - 5%)
OrHighMed 77.75 (2.7%) 77.85 (2.7%)
0.1% ( -5% - 5%)
SpanNear 10.66 (1.2%) 10.68 (1.2%)
0.2% ( -2% - 2%)
Phrase 59.75 (4.9%) 59.91 (5.2%)
0.3% ( -9% - 10%)
Term 1358.87 (6.8%) 1363.02 (6.1%)
0.3% ( -11% - 14%)
AndMedOrHighHigh 28.18 (3.0%) 28.27 (2.5%)
0.3% ( -5% - 6%)
OrHighHigh 18.55 (3.2%) 18.61 (2.2%)
0.3% ( -4% - 5%)
SloppyPhrase 19.41 (3.9%) 19.49 (3.5%)
0.4% ( -6% - 8%)
AndHighMed 65.81 (2.8%) 66.15 (2.4%)
0.5% ( -4% - 5%)
AndHighOrMedMed 36.49 (2.5%) 36.69 (1.9%)
0.5% ( -3% - 5%)
TermGroup100 12.19 (3.9%) 12.27 (4.0%)
0.6% ( -7% - 8%)
PKLookup 217.61 (3.2%) 220.39 (3.3%)
1.3% ( -5% - 8%)
Prefix3 197.95 (3.3%) 202.32 (3.4%)
2.2% ( -4% - 9%)
Wildcard 37.78 (2.2%) 41.43 (2.8%)
9.6% ( 4% - 14%)
Fuzzy1 47.77 (5.5%) 53.35 (8.4%)
11.7% ( -2% - 27%)
Fuzzy2 43.69 (7.5%) 49.50 (10.7%)
13.3% ( -4% - 34%)
Respell 34.05 (1.6%) 41.94 (1.4%)
23.2% ( 19% - 26%)
{noformat}
I plan to commit it and see how that affects nigthly benchmarks.
> Terms dictionary compression
> ----------------------------
>
> Key: LUCENE-4702
> URL: https://issues.apache.org/jira/browse/LUCENE-4702
> Project: Lucene - Core
> Issue Type: Wish
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Trivial
> Attachments: LUCENE-4702.patch, LUCENE-4702.patch
>
> Time Spent: 3h 40m
> Remaining Estimate: 0h
>
> I've done a quick test with the block tree terms dictionary by replacing a
> call to IndexOutput.writeBytes to write suffix bytes with a call to
> LZ4.compressHC to test the peformance hit. Interestingly, search performance
> was very good (see comparison table below) and the tim files were 14% smaller
> (from 150432 bytes overall to 129516).
> {noformat}
> TaskQPS baseline StdDevQPS compressed StdDev
> Pct diff
> Fuzzy1 111.50 (2.0%) 78.78 (1.5%)
> -29.4% ( -32% - -26%)
> Fuzzy2 36.99 (2.7%) 28.59 (1.5%)
> -22.7% ( -26% - -18%)
> Respell 122.86 (2.1%) 103.89 (1.7%)
> -15.4% ( -18% - -11%)
> Wildcard 100.58 (4.3%) 94.42 (3.2%)
> -6.1% ( -13% - 1%)
> Prefix3 124.90 (5.7%) 122.67 (4.7%)
> -1.8% ( -11% - 9%)
> OrHighLow 169.87 (6.8%) 167.77 (8.0%)
> -1.2% ( -15% - 14%)
> LowTerm 1949.85 (4.5%) 1929.02 (3.4%)
> -1.1% ( -8% - 7%)
> AndHighLow 2011.95 (3.5%) 1991.85 (3.3%)
> -1.0% ( -7% - 5%)
> OrHighHigh 155.63 (6.7%) 154.12 (7.9%)
> -1.0% ( -14% - 14%)
> AndHighHigh 341.82 (1.2%) 339.49 (1.7%)
> -0.7% ( -3% - 2%)
> OrHighMed 217.55 (6.3%) 216.16 (7.1%)
> -0.6% ( -13% - 13%)
> IntNRQ 53.10 (10.9%) 52.90 (8.6%)
> -0.4% ( -17% - 21%)
> MedTerm 998.11 (3.8%) 994.82 (5.6%)
> -0.3% ( -9% - 9%)
> MedSpanNear 60.50 (3.7%) 60.36 (4.8%)
> -0.2% ( -8% - 8%)
> HighSpanNear 19.74 (4.5%) 19.72 (5.1%)
> -0.1% ( -9% - 9%)
> LowSpanNear 101.93 (3.2%) 101.82 (4.4%)
> -0.1% ( -7% - 7%)
> AndHighMed 366.18 (1.7%) 366.93 (1.7%)
> 0.2% ( -3% - 3%)
> PKLookup 237.28 (4.0%) 237.96 (4.2%)
> 0.3% ( -7% - 8%)
> MedPhrase 173.17 (4.7%) 174.69 (4.7%)
> 0.9% ( -8% - 10%)
> LowSloppyPhrase 180.91 (2.6%) 182.79 (2.7%)
> 1.0% ( -4% - 6%)
> LowPhrase 374.64 (5.5%) 379.11 (5.8%)
> 1.2% ( -9% - 13%)
> HighTerm 253.14 (7.9%) 256.97 (11.4%)
> 1.5% ( -16% - 22%)
> HighPhrase 19.52 (10.6%) 19.83 (11.0%)
> 1.6% ( -18% - 25%)
> MedSloppyPhrase 141.90 (2.6%) 144.11 (2.5%)
> 1.6% ( -3% - 6%)
> HighSloppyPhrase 25.26 (4.8%) 25.97 (5.0%)
> 2.8% ( -6% - 13%)
> {noformat}
> Only queries which are very terms-dictionary-intensive got a performance hit
> (Fuzzy, Fuzzy2, Respell, Wildcard), other queries including Prefix3 behaved
> (surprisingly) well.
> Do you think of it as something worth exploring?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]