Transient Effects Case Study

Aus SDQ-Wiki
Zur Navigation springen Zur Suche springen

This page gives an overview of the case study investigating the effect of transient effects on the QoS of self-adaptive software system. The system under analysis was the Media Store reference media hosting application. We extended the Media Store application to support horizontal scaling in an OpenStack IaaS cloud environment. Documentation of the basis implementation and models of Media Store v3 extended for this case study can be found here.

Java EE implementation

The extended horizontally Media Store application can be found here. For information on the deployment in an OpenStack IaaS cloud, please contact Christian Stier.

Palladio Model

The analyzed PCM architecture model of the horizontally scaling Media Store application is available via SVN.

SimuLizar Setup

For instructions on how to set up SimuLizar, please refer to this page. The Action Model Interpreter and Actions meta-model has already been integrated with the SimuLizar trunk. However, we employed an branch of SimuLizar in the case study that uses aggregation functions not yet supported by the nightly SimuLizar version. The branch can be found in our SVN.

Load Driver

We used the extended JMeter 2.11 load driver combined with the TimeStamp timer plugin to execute timestamp-based load traces. The timestamp files used in our experiments are available for download. fourthscenarioTimeStamps.txt is the timestamp file of the workload used in Experiment A, fifthscenarioTimeStamps.txt for Experiment B and the third experiment.

Measurement Results

Both simulation and raw measurement logs can be found here. The archive also contains the R files used in the statistical analysis.

Directories containing "Fourth" refer results or predictions from Experiment A. The folder "ResultsFourth" contains the R scripts used in the analysis.

Directories containing "Fifth" refer to results from Experiment B. "ResultsV5" contains the R files used in the statistical analysis and for creating the plots.

"Refined" refers to the results of the third experiment, which refines the scalability rule used in Experiment B. "ResultsRefined" contains the R analysis scripts used for creating the boxplots.