srowen commented on a change in pull request #25992: [SPARK-29320][TESTS] Compare `sql/core` module in JDK8/11 (Part 1) URL: https://github.com/apache/spark/pull/25992#discussion_r330169048
########## File path: sql/core/benchmarks/ColumnarBatchBenchmark-jdk11-results.txt ########## @@ -0,0 +1,59 @@ +================================================================================================ +Int Read/Write +================================================================================================ + +OpenJDK 64-Bit Server VM 11.0.4+11-LTS on Linux 3.10.0-862.3.2.el7.x86_64 +Intel(R) Xeon(R) CPU E5-2670 v2 @ 2.50GHz +Int Read/Write: Best Time(ms) Avg Time(ms) Stdev(ms) Rate(M/s) Per Row(ns) Relative +------------------------------------------------------------------------------------------------------------------------ +Java Array 232 234 1 1409.7 0.7 1.0X +ByteBuffer Unsafe 396 396 0 827.9 1.2 0.6X +ByteBuffer API 602 603 0 544.0 1.8 0.4X +DirectByteBuffer 1185 1186 2 276.6 3.6 0.2X Review comment: Hm, I don't know. The cleaner hack that works in JDK 8 only works around an internal JVM check on max off heap limit. I would not imagine it causes slower performance, just may mean a user has to increase the off heap limit manually in extreme cases. That said, who knows what all the subtle implications are. It could be some other cause too, of course, simply some difference in JDK behavior. I don't think we want to force `--add-opens` or require people to use it of course, so I'd not benchmark that way. It's a valid observation in the benchmark, at least, to note that something seems to make it slower in JDK 11. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
