metaphori / experiments-neighbours-calculus

Computation Against a Neighbour

This is the project repository for the case study presented in a homonym manuscript submitted to a journal.

Structure of the repository

The project has the standard structure of Maven/Gradle projects.

• ./build.gradle contains the build definition and the tasks for compiling and packaging the application
• ./src/main/scala/it/unibo/casestudy/AdHocCloud.scala contains the source code of the case study
• ./src/main/yaml/adhoccloud.yml contains the configuration of the Alchemist simulation
• ./plotter.py is a script that can be used to plot data produced by Alchemist
• ./history/ contains, for different simulations, the corresponding configuration files

How-to

A simulation can be started with a command as the following:

$ ./gradlew --no-daemon && \
  java -Xmx5024m -cp "build/classes/main:libs/alchemist-redist-7.0.2-beta.jar:libs/scafi-lib-assembly-0.3.0.jar:build/resources/main" \
  it.unibo.alchemist.Alchemist \
  -b -var random leaderFailureProbability \
  -y src/main/yaml/adhoccloud.yml -e data/20181010_exp -t 400 -p 3 -v &> exec.txt &

where

• ./gradlew --no-daemon is used to compile and build the simulation source code
• java -cp "..." it.unibo.alchemist.Alchemist sets the java classpath and the main class to run
• -b -var <var1> <var2> sets batch mode, and the given variables are combined to form a matrix of simulation configurations to run
• -y <file> specifies the configuration for the simulation to run (which describes variables, i.e., input parameters, and exports)

• -e <basefilepath> specifies the base filepath for generating data output files
• -t 400 specifies the logical time to be simulated; -p <N> specifies the number of parallel simulations to run;
• -v &> exec.txt & sets the logging level as verbose and redirect output to a text file (e.g., for debugging purposes) and starts the task in background

Once data files (which are in CSV format) are generated, you can use script plotter.py to create plots.

# ./plotter.py <dir> <baseFilepath> <plotConfigFile>
$ ./plotter.py data 20181010_exp plotconfig.yml

where plotconfig.yml is something as the following:

the_plots_labels: # these should map the 'export fields'
  - Time
  - Available CPU
  - Actual CPU usage
  - Estimated CPU usage
  - Nfailures
  - Nleaders
the_plots_formats: # the plots to be created, and which fields to be selected
  - [0,2,3,4,5]    # the first field is used for X-axis; the others appear as curves
line_widths:
  - [1,1,1,1,1,1]
default_colors: &default_colors ["black","blue", "red","darkgreen","orange","violet"]
the_plots_colors:
  - *default_colors
legend_position:
  - "lower right"
sampling: True # when true, you use a 'random' (batch) dimension to unify multiple files and averaging the curves

Reproducing the experiments

NOTE: this may take a few hours.

In the paper, Fig. 12 shows, for different leaderProbabilityFailure values, the estimated CPU usage against the actual CPU usage, taking the mean values of 30 runs with varying random seeds. These plots can be re-generated with the following commands:

$ git fetch --all --tags --prune
$ git checkout tags/fig12 -b fig12

$ ./gradlew --no-daemon && \
 java -Xmx5024m -cp "build/classes/main:libs/alchemist-redist-7.0.2-beta.jar:libs/scafi-lib-assembly-0.3.0.jar:build/resources/main" \
 it.unibo.alchemist.Alchemist \
 -b -var leaderFailureProbability random \
 -y history/fig12/adhoccloud.yml \
 -e data/fig12 -t 850 -p 3 -v &> exec.txt &

$ ./plotter.py data fig12 history/fig12/plot.yml

Contacts

• Roberto Casadei: roby [dot] casadei [at] unibo [dot] it

Links

• ScaFi repository
• ScaFi website