[
https://issues.apache.org/jira/browse/LUCENE-10033?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17405002#comment-17405002
]
weizijun commented on LUCENE-10033:
-----------------------------------
Hi, [~jpountz], [~gsmiller], I run luceneutil: python3 src/python/localrun.py
-source wikimedium10k. I change some code in localrun.py:
{code:java}
import competition
import sys
# simple example that runs benchmark with WIKI_MEDIUM source and taks files
# Baseline here is ../lucene_baseline versus ../lucene_candidate
if __name__ == '__main__':
sourceData = competition.sourceData()
comp = competition.Competition()
index = comp.newIndex('lucene_baseline', sourceData,
facets = (('taxonomy:Date', 'Date'),
('taxonomy:Month', 'Month'),
('taxonomy:DayOfYear', 'DayOfYear'),
('sortedset:Month', 'Month'),
('sortedset:DayOfYear', 'DayOfYear')))
index_candidate = comp.newIndex('lucene_candidate', sourceData,
facets = (('taxonomy:Date', 'Date'),
('taxonomy:Month', 'Month'),
('taxonomy:DayOfYear', 'DayOfYear'),
('sortedset:Month', 'Month'),
('sortedset:DayOfYear', 'DayOfYear')))
#Warning -- Do not break the order of arguments
#TODO -- Fix the following by using argparser
if len(sys.argv) > 3 and sys.argv[3] == '-concurrentSearches':
concurrentSearches = True
else:
concurrentSearches = False
# create a competitor named baseline with sources in the ../trunk folder
comp.competitor('baseline', 'lucene_baseline',
index = index, concurrentSearches = concurrentSearches)
# use the same index here
# create a competitor named my_modified_version with sources in the ../patch
folder
# note that we haven't specified an index here, luceneutil will automatically
use the index from the base competitor for searching
# while the codec that is used for running this competitor is taken from this
competitor.
comp.competitor('my_modified_version', 'lucene_candidate',
index = index_candidate, concurrentSearches =
concurrentSearches)
# start the benchmark - this can take long depending on your index and
machines
comp.benchmark("baseline_vs_patch")
{code}
The baseline is lucene's master branch. The candidate is the branch from [PR
#1|https://github.com/jpountz/lucene/pull/1].
Here is the result:
{noformat}
TaskQPS baseline StdDevQPS my_modified_version
StdDev Pct diff p-value
BrowseMonthSSDVFacets 1958.41 (3.1%) 1538.44 (1.7%)
-21.4% ( -25% - -17%) 0.000
BrowseDayOfYearSSDVFacets 1652.84 (2.9%) 1413.20 (1.9%)
-14.5% ( -18% - -10%) 0.000
IntNRQ 1510.90 (4.4%) 1481.97 (8.3%)
-1.9% ( -13% - 11%) 0.361
HighIntervalsOrdered 557.92 (13.2%) 547.32 (12.6%)
-1.9% ( -24% - 27%) 0.642
MedIntervalsOrdered 872.92 (12.3%) 858.58 (11.6%)
-1.6% ( -22% - 25%) 0.664
LowIntervalsOrdered 1121.80 (6.3%) 1107.10 (6.7%)
-1.3% ( -13% - 12%) 0.526
MedPhrase 530.72 (5.8%) 524.26 (5.8%)
-1.2% ( -12% - 11%) 0.507
MedSpanNear 723.07 (3.8%) 714.34 (3.9%)
-1.2% ( -8% - 6%) 0.324
LowSloppyPhrase 942.46 (2.9%) 936.04 (3.4%)
-0.7% ( -6% - 5%) 0.497
LowPhrase 1131.13 (4.0%) 1128.82 (3.1%)
-0.2% ( -7% - 7%) 0.857
OrHighMed 655.21 (13.7%) 655.99 (12.2%)
0.1% ( -22% - 30%) 0.977
PKLookup 229.67 (1.6%) 230.10 (2.0%)
0.2% ( -3% - 3%) 0.754
OrHighLow 634.01 (10.4%) 635.60 (5.9%)
0.3% ( -14% - 18%) 0.925
HighTerm 3600.11 (5.8%) 3611.93 (4.3%)
0.3% ( -9% - 11%) 0.839
HighSloppyPhrase 367.37 (5.1%) 368.73 (5.8%)
0.4% ( -10% - 11%) 0.832
HighSpanNear 421.73 (6.5%) 423.96 (5.8%)
0.5% ( -11% - 13%) 0.787
HighTermDayOfYearSort 2533.62 (7.7%) 2549.91 (7.3%)
0.6% ( -13% - 16%) 0.786
LowSpanNear 497.84 (5.5%) 502.07 (3.3%)
0.8% ( -7% - 10%) 0.553
Respell 266.07 (12.8%) 268.61 (12.2%)
1.0% ( -21% - 29%) 0.809
HighPhrase 622.36 (6.2%) 629.01 (7.7%)
1.1% ( -12% - 15%) 0.629
AndHighMed 854.35 (5.2%) 865.51 (3.7%)
1.3% ( -7% - 10%) 0.360
BrowseMonthTaxoFacets 3057.03 (5.9%) 3097.61 (4.9%)
1.3% ( -8% - 12%) 0.436
BrowseDayOfYearTaxoFacets 2399.39 (5.0%) 2432.18 (4.0%)
1.4% ( -7% - 10%) 0.336
HighTermMonthSort 2564.47 (6.1%) 2607.36 (4.7%)
1.7% ( -8% - 13%) 0.330
Fuzzy1 306.10 (7.1%) 311.26 (7.0%)
1.7% ( -11% - 17%) 0.451
LowTerm 3912.29 (4.3%) 3979.32 (6.2%)
1.7% ( -8% - 12%) 0.309
OrHighHigh 480.12 (7.7%) 488.87 (7.5%)
1.8% ( -12% - 18%) 0.447
Prefix3 471.26 (15.0%) 480.73 (15.7%)
2.0% ( -24% - 38%) 0.679
BrowseDateTaxoFacets 2721.15 (4.8%) 2777.53 (4.5%)
2.1% ( -6% - 12%) 0.163
AndHighHigh 1023.76 (7.7%) 1045.20 (7.3%)
2.1% ( -11% - 18%) 0.377
MedTerm 3813.24 (5.5%) 3898.93 (5.5%)
2.2% ( -8% - 14%) 0.198
Fuzzy2 102.35 (12.7%) 104.67 (15.1%)
2.3% ( -22% - 34%) 0.608
AndHighLow 3004.48 (5.9%) 3073.96 (6.7%)
2.3% ( -9% - 15%) 0.246
MedSloppyPhrase 591.31 (4.2%) 605.55 (3.6%)
2.4% ( -5% - 10%) 0.050
Wildcard 544.95 (12.9%) 577.08 (7.2%)
5.9% ( -12% - 29%) 0.074
{noformat}
And the whole result is from the Attachment: [^benchmark]
> Encode doc values in smaller blocks of values, like postings
> ------------------------------------------------------------
>
> Key: LUCENE-10033
> URL: https://issues.apache.org/jira/browse/LUCENE-10033
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Priority: Minor
> Attachments: benchmark
>
> Time Spent: 1h
> Remaining Estimate: 0h
>
> This is a follow-up to the discussion on this thread:
> https://lists.apache.org/thread.html/r7b757074d5f02874ce3a295b0007dff486bc10d08fb0b5e5a4ba72c5%40%3Cdev.lucene.apache.org%3E.
> Our current approach for doc values uses large blocks of 16k values where
> values can be decompressed independently, using DirectWriter/DirectReader.
> This is a bit inefficient in some cases, e.g. a single outlier can grow the
> number of bits per value for the entire block, we can't easily use run-length
> compression, etc. Plus, it encourages using a different sub-class for every
> compression technique, which puts pressure on the JVM.
> We'd like to move to an approach that would be more similar to postings with
> smaller blocks (e.g. 128 values) whose values get all decompressed at once
> (using SIMD instructions), with skip data within blocks in order to
> efficiently skip to arbitrary doc IDs (or maybe still use jump tables as
> today's doc values, and as discussed here for postings:
> https://lists.apache.org/thread.html/r7c3cb7ab143fd4ecbc05c04064d10ef9fb50c5b4d6479b0f35732677%40%3Cdev.lucene.apache.org%3E).
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
