gf2121 opened a new issue, #11891: URL: https://github.com/apache/lucene/issues/11891
### Description In https://github.com/apache/lucene/issues/11202 we introduced a MergeInstance of DocValuesProducer to speed up merge by bulk decoding. If i understood correctly, MergeInstance should run faster than `DirectPackedReader` when value reading is dense. So I tried to use MergeInstance to speed up `FastTaxonomyFacetCounts#countAll` like this: ``` LeafReader leafReader = context.reader(); SortedNumericDocValues multiValued; if (leafReader instanceof CodecReader) { FieldInfo fieldInfo = leafReader.getFieldInfos().fieldInfo(indexFieldName); multiValued = ((CodecReader) leafReader).getDocValuesReader().getMergeInstance().getSortedNumeric(fieldInfo); } else { multiValued = context.reader().getSortedNumericDocValues(indexFieldName); } ... ``` Results is a bit disappointing: ``` TaskQPS baseline StdDevQPS my_modified_version StdDev Pct diff p-value BrowseMonthTaxoFacets 11.66 (10.3%) 4.82 (6.9%) -58.7% ( -68% - -46%) 0.000 BrowseRandomLabelTaxoFacets 6.96 (15.6%) 4.18 (11.8%) -39.9% ( -58% - -14%) 0.000 BrowseDateTaxoFacets 9.96 (10.1%) 6.13 (12.2%) -38.5% ( -55% - -17%) 0.000 BrowseDayOfYearTaxoFacets 10.17 (8.5%) 6.46 (9.3%) -36.5% ( -50% - -20%) 0.000 ``` I wonder if MergeInstance can run slower for some bpv even if the value reading is dense? Maybe we can optimize for that? Forgive me if i misunderstood something :) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
