Major Description
Applying the same approach as LUCENE-4923 gives ~ 30% improvement according to luceneutil.
Task QPS trunk StdDev QPS patch StdDev Pct diff
OrMaxHighHigh 17.94 (2.7%) 23.23 (3.1%) 29.5% ( 23% - 36%)
OrMaxHighLow 28.08 (2.6%) 37.11 (3.9%) 32.1% ( 25% - 39%)
OrMaxHighMed 16.80 (2.7%) 22.25 (3.9%) 32.5% ( 25% - 40%)
diff -r e0ea40cf243c perf/TaskParser.java
--- a/perf/TaskParser.java Mon Apr 01 14:49:08 2013 +0200
+++ b/perf/TaskParser.java Wed Apr 10 22:35:40 2013 -0400
@@ -31,6 +31,7 @@
import org.apache.lucene.search.BooleanClause;
import org.apache.lucene.search.BooleanQuery;
import org.apache.lucene.search.CachingWrapperFilter;
+import org.apache.lucene.search.DisjunctionMaxQuery;
import org.apache.lucene.search.Filter;
import org.apache.lucene.search.NumericRangeQuery;
import org.apache.lucene.search.Query;
@@ -169,6 +170,17 @@
true);
sort = null;
group = null;
+ } else if (text.startsWith("disjunctionMax//")) {
+ final int spot3 = text.indexOf(' ');
+ if (spot3 == -1) {
+ throw new RuntimeException("failed to parse query=" + text);
+ }
+ DisjunctionMaxQuery dismax = new DisjunctionMaxQuery(1f);
+ dismax.add(new TermQuery(new Term(fieldName, text.substring(16, spot3))));
+ dismax.add(new TermQuery(new Term(fieldName, text.substring(spot3+1).trim())));
+ query = dismax;
+ sort = null;
+ group = null;
} else if (text.startsWith("nrq//")) {
// field start end
final int spot3 = text.indexOf(' ');
cat wikimedium.10M.nostopwords.tasks | grep "^Or" | sed -e "s/Or\([a-zA-Z]*\)\:\ /OrMax\1\:\ disjunctionMax\/\//g" > dismax.tasks