Branch: master
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
src/main
README.md
pom.xml

README.md

JMH Performance tests

Build

To build the benchmark iteration tests:

$ mvn clean install

Run it from command line using an "uber jar"

There is an alternative way to run it which might be handy when needing to copy the performance test to a dedicated server and for some reason you're not liking rsync:

$ java -jar target/benchmarks.jar

Run flight recorder on Infinispan with many iterations

$ java -jar target/benchmarks.jar -jvmArgsPrepend "-XX:+UnlockCommercialFeatures -XX:+FlightRecorder -XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints -XX:FlightRecorderOptions=stackdepth=128" -e NoRehash -i 500 -t 2

  • jvmArgsPrepend: allows to add additional JVM properties to the forked process which runs the benchmark
  • e: excludes running all tests including this name (for example ignore tests that don't have rehash enabled)
  • i: sets the number of iterations. Setting it very high here to have time to play with Flight Recorder before it shuts down
  • t: sets the number of worker threads. If you wish to test concurrent iteration requests

Set benchmark parameters

Benchmark parameters (those who use @Param) will get the benchmark repeated to satisfy each possible permutation. Sometimes you want to limit execution to a specific condition, e.g.:

-p useStrings=false

Will only use objects for the payload and skip the test run using strings. Strings are quite a bit slower.

For all possible paramters see InfinispanHolder and KeySequenceGenerator

Start a cluster node without the benchmark

Infinispan node; set the appropriate interface address; in this case you can use a pattern: $ java -cp target/benchmarks.jar -Dbind_address=match-address:192.168.* -Djava.net.preferIPv4Stack=true org.infinispan.jmhbenchmarks.InfinispanNodeStarter

This needs to be added still.

Start a distributed benchmark

To run the benchmark and have it connect to other nodes on the cluster but running "standalone" (see previous paragraph), set the 'distributed=true' flag.

For example to run all Hazelcast test benchmarks while connecting to other nodes on the cluster:

$ java -jar target/benchmarks.jar -jvmArgsPrepend "-Dbind_address=192.168.1.20 -Djava.net.preferIPv4Stack=true -Ddistributed=true" -e getPutInfinispan -e infinispanG -e infinispanP

Notes

For best results disable features such as power management, dynamic CPU scaling, and run it on a dedicated box which has no other significant services running. So the "run it from your IDE" approach is just meant for development of new tests.

Network checklist

  • Disable any firewall! e.g. iptables

Make sure you have a route for the 232.0.0.0 routed to the network interface connected to the cluster:

Sample route output: 232.0.0.0 0.0.0.0 248.0.0.0 U 0 0 0 enp0s25

TODO

  • make sure there's some key overlap between the threads (and unit tests for that?)
  • have it run the other nodes in physically different boxes
  • improve scenario parameters