Benchmarks for TDSC
- author: Fabien Hermenier
- contact: firstname.lastname@example.org
This directory contains everything is needed to replay micro benchmarks of BtrPlace.
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
benchmarks-tdsc-VERSION.tar.gz will be available in the
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.
dispatcher.sh read the RPs on the workload and will distribute them on the slaves running the
application. Launch them to get their parameters.
High-level scripts (
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
The following command run all the benchmarks. It uses
the workload inside
wkld-tdsc and stores the result inside the folder
$ ./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
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
$ ./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
to generate a workload, use
$ ./generate.sh 10 simple_workload
to compute the number of applications impacted by a load increase in a given workload, use
$ ./impacted.sh simple_workload/r3 c100p5000