WeakOrderQueue.transfer is 1st Hot Method

Expected behavior

4.1.44.Final should work same as 4.0.49.Final(WeakOrderQueue#transfer) Hot threads and methods in 4.0.49

Actual behavior

4.0.49.Final works better than 4.1.44.Final Hot threads and methods in 4.1.44

Steps to reproduce

Be part of a service, it is hard to strip a demo code.

Netty version

4.1.44.Final

JVM version (e.g. java -version)

java version “1.8.0_171” Java™ SE Runtime Environment (build 1.8.0_171-b11) Java HotSpot™ 64-Bit Server VM (build 25.171-b11, mixed mode)

OS version (e.g. uname -a)

Linux 3.10.0-957.21.3.el7.x86_64 #1 SMP Tue Jun 18 16:35:19 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

[franz1981]

@louxiu using -XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints is useful when inlined methods are the most costy ones: if it won’t make any difference “maybe” is that they are not the most costy one, but the flags should be there to make sure it’s the case

[louxiu]

@louxiu so just so I understand correct… There are a lot of “empty” WeakOrderQueue and the problem is the iteration of 64 links ?

Yes. Currently we disable drop ratio. It’s seems that only drop ratio above some threshold, it will be a problem. Not sure wether there is an easy way(or worth to optimize) to avoid loop “empty” WeakOrderQueue list.