Benchmarks for TDSC

• author: Fabien Hermenier
• contact: fabien.hermenier@unice.fr

This directory contains everything is needed to replay micro benchmarks of BtrPlace.

Compilation instruction

The application is managed by maven. You must install it to compile the benchmark. Once install, go it this directory and type:

$ mvn assembly:assembly

The distribution benchmarks-tdsc-VERSION.tar.gz will be available in the target directory.

Running benchmarks

This archive does not contain a workload but you can download the default one at this address. Once downloaded, un-compress it inside the application root directory.

Basic scripts are available to process each part of the workload. The script dispatcher.sh read the RPs on the workload and will distribute them on the slaves running the handler.sh application. Launch them to get their parameters.

High-level scripts (bench-all.sh, bench-ratio.sh, and bench-ratio-single.sh) made the processing of the workload easy. They all rely on a *-process.sh script to distribute the workload. By default, g5k-process.sh is used to run process the workload on the Grid'5000 platforms. local-process.sh allows the execution of the benchmark on a single machine. To set the process script, edit bench-ratio-single.sh.

The following command run all the benchmarks. It uses the workload inside wkld-tdsc and stores the result inside the folder output.

$ ./bench-all.sh  wkld-tdsc output

Executing this benchmark will take hours. Especially the benchmark without the filter option. Edit the scripts to adapt the benchmark to your needs.

Inside the output folder, Each sub-folder contains the resulting reconfiguration plans and a file called result.data that aggregates the results.

Exploiting the benchmarks results

To compress only the benchmark results:

$ tar cfz output-results.tar.gz output/*/result.data

To generate a HTML report of the benchmarks, use the following command to generate the output.html file:

$ ./report-HTML.pl output/* > output.html

To store in the directory datafiles the data that will be used to generate graphics:

$ ./make_datafiles.sh output datafiles

To generate the graphics inside the graphs folder, you have to install R then run the following command:

$ ./graphs.sh datafiles graphs

Additional tools

• to generate a workload, use generate.sh:

  $ ./generate.sh 10 simple_workload

• to compute the number of applications impacted by a load increase in a given workload, use impacted.sh:

  $ ./impacted.sh simple_workload/r3 c100p5000