mikemccand commented on issue #15662:
URL: https://github.com/apache/lucene/issues/15662#issuecomment-3842918834

   01/22 (good) run has these `perf stat` counters for one of the 20 
`SearchPerfTest` JVMs:
   
   ```
    Performance counter stats for '/usr/lib/jvm/java-25-openjdk/bin/java 
--add-modules jdk.incubator.vector -Xmx32g -Xms32g -server -XX:+UseParallelGC 
-Djava.io.tmpdir=/l/tmp --enable-native-access=ALL-UNNAMED -XX:StartFligh\
   
tRecording=dumponexit=true,maxsize=250M,settings=/l/util.nightly/src/python/profiling.jfc,filename=/l/logs.nightly/bench-search-nightly-nightly-9.jfr
 -XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints -classpath /l/tr\
   
unk.nightly/lucene/core/build/libs/lucene-core-11.0.0-SNAPSHOT.jar:/l/trunk.nightly/lucene/sandbox/build/classes/java/main:/l/trunk.nightly/lucene/misc/build/classes/java/main:/l/trunk.nightly/lucene/facet/build/classes/j\
   
ava/main:/l/trunk.nightly/lucene/analysis/common/build/classes/java/main:/l/trunk.nightly/lucene/analysis/icu/build/classes/java/main:/l/trunk.nightly/lucene/queryparser/build/classes/java/main:/l/trunk.nightly/lucene/gro\
   
uping/build/classes/java/main:/l/trunk.nightly/lucene/suggest/build/classes/java/main:/l/trunk.nightly/lucene/highlighter/build/classes/java/main:/l/trunk.nightly/lucene/codecs/build/classes/java/main:/l/trunk.nightly/luc\
   
ene/queries/build/classes/java/main:/l/trunk.nightly/lucene/join/build/classes/java/main:/l/trunk.nightly/lucene/spatial3d/build/classes/java/main:/l/util.nightly/lib/HdrHistogram.jar:/l/util.nightly/src/main/build/classe\
   s/java/main:/l/util.nightly/build perf.SearchPerfTest -dirImpl MMapDirectory 
-indexPath 
/l/indices/wikimedium.trunk.nightly.facets.taxonomy:Date.taxonomy:Month.taxonomy:DayOfYear.sortedset:Month.sortedset:DayOfYear.sorted\
   
set:Date.taxonomy:RandomLabel.sortedset:RandomLabel.Lucene90.Lucene104.dvfields.vectors=1024.nd27.625M
 -facets taxonomy:Date;Date -facets taxonomy:Month;Month -facets 
taxonomy:DayOfYear;DayOfYear -facets sortedset:Month;M\
   onth -facets sortedset:DayOfYear;DayOfYear -facets sortedset:Date;Date 
-facets taxonomy:RandomLabel;RandomLabel -facets 
sortedset:RandomLabel;RandomLabel -analyzer StandardAnalyzerNoStopWords 
-taskSource /l/util.nightly/t\
   asks/wikinightly.tasks -numConcurrentQueries 1 -taskRepeatCount 50 -field 
body -tasksPerCat 5 -searchConcurrency 8 -staticSeed -8035476 -seed -7082989 
-similarity BM25Similarity -commit multi -hiliteImpl FastVectorHighlig\
   hter -log /l/logs.nightly/nightly.nightly.9 -topN 100 -context  -printHeap 
-pk -vectorFile 
/lucenedata/enwiki/cohere-v3/cohere-v3-wikipedia-en-scattered-1024d.queries.vec 
-vectorDimension 1024 -pollute':
   
         1,088,849.85 msec task-clock:u                     #    2.728 CPUs 
utilized
                    0      context-switches:u               #    0.000 /sec
                    0      cpu-migrations:u                 #    0.000 /sec
              603,814      page-faults:u                    #  554.543 /sec
    9,322,771,152,253      instructions:u                   #    2.13  insn per 
cycle
                                                     #    0.04  stalled cycles 
per insn     (30.78%)
    4,379,914,648,276      cycles:u                         #    4.023 GHz      
                   (30.77%)
      398,139,212,961      stalled-cycles-frontend:u        #    9.09% frontend 
cycles idle        (30.76%)
    1,777,065,993,098      branches:u                       #    1.632 G/sec    
                   (30.76%)
       23,633,297,198      branch-misses:u                  #    1.33% of all 
branches             (30.76%)
    4,403,556,999,556      L1-dcache-loads:u                #    4.044 G/sec    
                   (30.76%)
       39,427,726,537      L1-dcache-load-misses:u          #    0.90% of all 
L1-dcache accesses   (30.77%)
      622,042,664,608      L1-icache-loads:u                #  571.284 M/sec    
                   (30.77%)
        1,693,096,314      L1-icache-load-misses:u          #    0.27% of all 
L1-icache accesses   (30.77%)
        1,829,585,485      dTLB-loads:u                     #    1.680 M/sec    
                   (30.77%)
          371,109,765      dTLB-load-misses:u               #   20.28% of all 
dTLB cache accesses  (30.77%)
          214,576,217      iTLB-loads:u                     #  197.067 K/sec    
                   (30.78%)
           23,720,429      iTLB-load-misses:u               #   11.05% of all 
iTLB cache accesses  (30.78%)
   
        399.117771506 seconds time elapsed
   
       1035.914308000 seconds user
         42.341318000 seconds sys
   ```
   
   1/29 (bad) has these:
   
   ```
    Performance counter stats for '/usr/lib/jvm/java-25-openjdk/bin/java 
--add-modules jdk.incubator.vector -Xmx32g -Xms32g -server -XX:+UseParallelGC 
-Djava.io.tmpdir=/l/tmp --enable-native-access=ALL-UNNAMED -XX:StartFligh\
   
tRecording=dumponexit=true,maxsize=250M,settings=/l/util.nightly/src/python/profiling.jfc,filename=/l/logs.nightly/bench-search-nightly-nightly-9.jfr
 -XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints -classpath /l/tr\
   
unk.nightly/lucene/core/build/libs/lucene-core-11.0.0-SNAPSHOT.jar:/l/trunk.nightly/lucene/sandbox/build/classes/java/main:/l/trunk.nightly/lucene/misc/build/classes/java/main:/l/trunk.nightly/lucene/facet/build/classes/j\
   
ava/main:/l/trunk.nightly/lucene/analysis/common/build/classes/java/main:/l/trunk.nightly/lucene/analysis/icu/build/classes/java/main:/l/trunk.nightly/lucene/queryparser/build/classes/java/main:/l/trunk.nightly/lucene/gro\
   
uping/build/classes/java/main:/l/trunk.nightly/lucene/suggest/build/classes/java/main:/l/trunk.nightly/lucene/highlighter/build/classes/java/main:/l/trunk.nightly/lucene/codecs/build/classes/java/main:/l/trunk.nightly/luc\
   
ene/queries/build/classes/java/main:/l/trunk.nightly/lucene/join/build/classes/java/main:/l/trunk.nightly/lucene/spatial3d/build/classes/java/main:/l/util.nightly/lib/HdrHistogram.jar:/l/util.nightly/src/main/build/classe\
   s/java/main:/l/util.nightly/build perf.SearchPerfTest -dirImpl MMapDirectory 
-indexPath 
/l/indices/wikimedium.trunk.nightly.facets.taxonomy:Date.taxonomy:Month.taxonomy:DayOfYear.sortedset:Month.sortedset:DayOfYear.sorted\
   
set:Date.taxonomy:RandomLabel.sortedset:RandomLabel.Lucene90.Lucene104.dvfields.vectors=1024.nd27.625M
 -facets taxonomy:Date;Date -facets taxonomy:Month;Month -facets 
taxonomy:DayOfYear;DayOfYear -facets sortedset:Month;M\
   onth -facets sortedset:DayOfYear;DayOfYear -facets sortedset:Date;Date 
-facets taxonomy:RandomLabel;RandomLabel -facets 
sortedset:RandomLabel;RandomLabel -analyzer StandardAnalyzerNoStopWords 
-taskSource /l/util.nightly/t\
   asks/wikinightly.tasks -numConcurrentQueries 1 -taskRepeatCount 50 -field 
body -tasksPerCat 5 -searchConcurrency 8 -staticSeed -8035476 -seed -7082989 
-similarity BM25Similarity -commit multi -hiliteImpl FastVectorHighlig\
   hter -log /l/logs.nightly/nightly.nightly.9 -topN 100 -context  -printHeap 
-pk -vectorFile 
/lucenedata/enwiki/cohere-v3/cohere-v3-wikipedia-en-scattered-1024d.queries.vec 
-vectorDimension 1024 -pollute':
   
    1,566,239,675,863      task-clock:u                     #    2.851 CPUs 
utilized
                    0      context-switches:u               #    0.000 /sec
                    0      cpu-migrations:u                 #    0.000 /sec
              722,319      page-faults:u                    #  461.180 /sec
    9,378,561,411,799      instructions:u                   #    2.22  insn per 
cycle
                                                     #    0.04  stalled cycles 
per insn     (30.77%)
    4,228,223,018,839      cycles:u                         #    2.700 GHz      
                   (30.78%)
      393,219,880,128      stalled-cycles-frontend:u        #    9.30% frontend 
cycles idle        (30.77%)
    1,777,249,743,096      branches:u                       #    1.135 G/sec    
                   (30.78%)
       23,578,816,975      branch-misses:u                  #    1.33% of all 
branches             (30.77%)
    4,449,403,141,862      L1-dcache-loads:u                #    2.841 G/sec    
                   (30.77%)
       40,782,544,753      L1-dcache-load-misses:u          #    0.92% of all 
L1-dcache accesses   (30.77%)
      593,096,321,940      L1-icache-loads:u                #  378.675 M/sec    
                   (30.77%)
        1,713,082,414      L1-icache-load-misses:u          #    0.29% of all 
L1-icache accesses   (30.78%)
        1,908,659,533      dTLB-loads:u                     #    1.219 M/sec    
                   (30.79%)
          349,067,129      dTLB-load-misses:u               #   18.29% of all 
dTLB cache accesses  (30.78%)
          179,164,433      iTLB-loads:u                     #  114.391 K/sec    
                   (30.78%)
           21,113,672      iTLB-load-misses:u               #   11.78% of all 
iTLB cache accesses  (30.78%)
   
        549.354958865 seconds time elapsed
   
       1479.501112000 seconds user
         64.359562000 seconds sys
   ```
   
   I find the 4.023 GHz vs 2.700 GHz ("effective clock speed" of the CPU 
disturbing -- the box just ran fewer instructions per second overall.  Thermal 
throttling?  Different CPU governor?  But @msokolov pointed out that e.g. more 
blocking on IO, or thread lock contention, could also manifest this way maybe?


-- 
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: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to