JReferral

A tool that run your java program through multiple versions of JVMs and optimisation options to recommand the most energry efficiant configuration.

Requirements

• Docker : we use a docker image for testing and isolating the JVM
• Linux : Our tool is based on RAPL sensors to measure energy
• Python 3.5+ : for the reporting purpose

How to use

1. Choose The configurations to be tested in th file jvms.sh

   • Uncomment what should be included during the tests

2. Run build-images.sh that will build the docker images for respective jvms

   • You can add the option -u to setup the username for docker images
   • The images will have the format*username/jvm:version

3. Replace the word java with src/jrecommand.sh in your command

   • Example

     java -jar example.jar arg1 arg2 ...
     will be 
     ./jreferral.sh -jar example.jar arg1 arg2 ...
     

You will find the results in data.csv and the execution log in exp.log

Note:

Due to security reasons, you need sudo access in order to measure the energy

Bulk Benchmarking

If you want to test multiple benchmarks you can use bulk.sh, to do So

1. Put your benchmarks in a file as an exmple example_bulks.sh

2. Run the script bulk.sh giving him as entry the path of your bulks_file

   • Example

     ./src/bulk.sh example_bulks.sh
     

Plotting the results

With the option -p the program will generate a pdf file containaing all the measures aka execution time and energies for each benchmark.

• In order to have more visible name for the different tunning options you can put them in the tagging file using the following format

  options TAG visible tag

• Example:

  -XX:+UnlockExperimentalVMOptions -XX:+UseEpsilonGC -XX:+AlwaysPreTouch TAG EpsilonGC
  

The tests with this options will take the name of EpsilonGC in the graphs

Flags and options

• For both jreferral.sh and bulk.sh

Flag	Description	Default value
-u name	The username of docker images	user
-o filename	The filename where the all the measures should be put in	data.csv
-l filename	Redirection of the output of the tests	exp.log
-n Number	The number of iterations that should be run for each configuration	1
-s Duration	Time to wait between two consecutive tests (in order to avoid the test impacting each others )	3s
-d	print the DRAM energy and CPU energy seperately	False
-p	generate a pdf that containes all the measures in format of boxplot	False

Example

Recap

This is an example to test the energy consumption of zip4j

and for the detailed version (aka -d)

Plots

Energy consumption of CPU

Energy consumption of DRAM

Execution of Time