FW: Basic JMH questions

[Bernd Eckenfels]

BTW profiler options can be listed:

# java -jar benchmarks.jar -prof stack:help

Kenneth Fogel wrote on 16. Aug 2024 20:45 (GMT +02:00):

> (I may have used the wrong email address earlier today so sending this
> again)
> 
> Hi,
> 
> If there is a better mailing list to ask my questions about the JMH then
> please let me know. I will be presenting a session at JavaZone next month
> entitled "I Just Discovered the Java Microbenchmark Harness and You Should
> To". Some years ago, I used System.nanoTime to measure performance of the
> Java Collections for a course I was giving. The results were relatively
> correct, i.e. LinkedLists are slow,  but I knew they were far from
> accurate. They also changed radically from run to run. I did a similar
> presentation two years ago but for this version I wish to present what
> will effectively be a tutorial on the best use of the JMH. To that end I
> have just a few questions.
> 
> The first most obvious question is if there are JavaDocs. Experimentation
> has revealed to me much of how the JMH works. There are many good articles
> on the JMH, but none are complete which is what leads to the next two
> questions.
> 
> When is a new JVM spun up for testing? Does it happen for every fork,
> warmup, or iteration? In reviewing my results, I see that the first Warmup
> Iteration is usually twice what the subsequent warmup iterations are and
> that the Iteration values are usually all close together. I am using the
> default of 5 forks and iterations, both in 10s. I am using Java 21.
> 
> What does this option mean:
> .jvmArgsAppend("-Djmh.stack.period=1") mean
> I found it in this file at 
> https://github.com/xp-vit/java-logging-performance-tests in the
> BenchmarkUtils.java<https://github.com/xp-vit/java-logging-performance-tests%20in%20the%20BenchmarkUtils.java>.
> I do not perceive a difference if I use it or not.
> 
> Please feel free to recommend any topics or concepts that should be
> presented. The session is only an hour, but I fear I may already have
> enough for two or three hours. As running tests can be time consuming, I
> will show the results of tests I performed previously. A single test class
> with 23 methods took three and a half hours to run last night.
> 
> Here is the setup for each test running under Windows 11:
> # JMH version: 1.37
> # VM version: JDK 21.0.2, OpenJDK 64-Bit Server VM, 21.0.2+13-LTS
> # VM invoker: C:\Program Files\Eclipse
> Adoptium\jdk-21.0.2.13-hotspot\bin\java.exe
> # VM options: -server -Djmh.stack.period=1
> # Blackhole mode: compiler (auto-detected, use
> -Djmh.blackhole.autoDetect=false to disable)
> # Warmup: 5 iterations, 10 s each
> # Measurement: 5 iterations, 10 s each
> # Timeout: 10 min per iteration
> # Threads: 1 thread, will synchronize iterations
> # Benchmark mode: Average time, time/op
> 
> Thank you for your time.
> 
> Ken Fogel

Gruß
Bernd
— 
https://bernd.eckenfels.net