You are viewing a plain text version of this content. The canonical link for it is here.

Posted to notifications@geode.apache.org by "jhuynh1 (GitHub)" <gi...@apache.org> on 2019/02/22 22:56:35 UTC

[GitHub] [geode] jhuynh1 commented on issue #3196: GEODE-6404: work around possible sync issue with computeIfAbsent

Using the added benchmark on my local mac I had the following results:
Java 1.8
Old way (straight computeIfAbsent)
ComputeIfAbsentBenchmark.monitorQuery  thrpt   25  597,050.898 ± 23260.717  ops/s
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  851,289.826 ± 27422.826  ops/s

New way (doing get)
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  33,338,554.778 ± 2,533,816.655  ops/s
(Ran a few more times but forgot to log the results, they were consistent with being exponentially faster)

Java 11 (compiled against java 1.8)
Old way (straight computeIfAbsent)
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  17,292,751.710 ± 1,975,003.721  ops/s
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  19,575,401.472 ± 682464.230  ops/s
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  20,644,562.716 ± 511267.928  ops/s

New way (doing get)
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  32,063,248.194 ± 831,342.655  ops/s
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  31,672,072.486 ± 1095001.901  ops/s
ComputeIfAbsentBenchmark.monitorQuery  thrpt   10  32,283,562.515 ± 1837351.361  ops/s

It looks like jdk 11 has vastly improved computeIfAbsent but the retrieval of computeIfAbsent is slightly slower than the get still.   I've removed the check for the jdk version for now based on these results...

[ Full content available at: https://github.com/apache/geode/pull/3196 ]
This message was relayed via gitbox.apache.org for notifications@geode.apache.org