[ https://issues.apache.org/jira/browse/FLINK-35215?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17840230#comment-17840230 ]
Qinghui Xu commented on FLINK-35215: ------------------------------------ Hello [~fanrui] I have a few questions: * I'm trying to understand the numbers. How should I interprete them? ** For [serializerKryoWithoutRegistration|[http://flink-speed.xyz/timeline/#/?exe=1,6,12&ben=serializerKryoWithoutRegistration&extr=on&quarts=on&equid=off&env=3&revs=50]:|http://flink-speed.xyz/timeline/#/?exe=1,6,12&ben=serializerKryoWithoutRegistration&extr=on&quarts=on&equid=off&env=3&revs=50],] the number for java17 had dropped before my commit was merged, and was a bit better after; the number for default jdk (java8?) dropped as well before my commit, between 04-16 and 04-18, there's a drop about 5%, and then the drop between 04-18 and 04-20 is less than 2% (I skipped 04-19 because it looks like an outlier). ** For [serializerKryo|[http://flink-speed.xyz/timeline/#/?exe=1,6,12&ben=serializerKryo&extr=on&quarts=on&equid=off&env=3&revs=50],] the number for java8 seems like normal fluctuation if I compare with the past; for java17 the drop between 04-17 and 04-19 (before my commit) is around 5%, and numbers are not much changed after the merge of my commit comparing to 04-19. ** For the two dashboards, only java11 seems have dropped in a statistically meaningful way. But in the past, the fluctuation of this trend has always been significantly higher, so I'm not sure if it's really related to my commit. * How are the benchmarks done? Do you have a documentation to reproduce that? > The performance of serializerKryo and serializerKryoWithoutRegistration are > regressed > ------------------------------------------------------------------------------------- > > Key: FLINK-35215 > URL: https://issues.apache.org/jira/browse/FLINK-35215 > Project: Flink > Issue Type: Bug > Components: API / Type Serialization System > Affects Versions: 1.20.0 > Reporter: Rui Fan > Priority: Blocker > > The performance of serializerKryo and serializerKryoWithoutRegistration are > regressed[1][2], I checked recent commits, and found FLINK-34954 changed > related logic. > > [1] > [http://flink-speed.xyz/timeline/#/?exe=1,6,12&ben=serializerKryo&extr=on&quarts=on&equid=off&env=3&revs=50] > [2] > http://flink-speed.xyz/timeline/#/?exe=1,6,12&ben=serializerKryoWithoutRegistration&extr=on&quarts=on&equid=off&env=3&revs=50 > > -- This message was sent by Atlassian Jira (v8.20.10#820010)