van Hoorn, Andre, Vögele, Christian, Schulz, Eike, Hasselbring, Wilhelm and Krcmar, Helmut (2014) Integrating Workload Specification and Extraction for Model-Based and Measurement-Based Performance Evaluation: An Approach for Session-Based Software Systems [Talk] In: Symposium on Software Performance 2014: Joint Descartes/Kieker/Palladio Days, November 26-28, 2014, Stuttgart, Germany.
20141127_SOSP14_Final.pdf - Presentation
Download (2114Kb) | Preview
Workload modeling and generation/simulation are essential to systematically and accurately evaluate performance properties of software systems for both measurement-based (e.g., load testing and benchmarking) and model-based performance evaluation (e.g., prediction). For measurement-based evaluation, established load generation tools exist that take workload specifications as input and generate corresponding requests to the system under test (SUT). Specifications of workloads are also tightly integrated in formalisms and approaches for model-based performance evaluation, including analytic and simulative techniques. The definition of workload specifications that represent the real workload as accurately as possible is one of the biggest challenges in both areas. Only few approaches for sharing workload specifications between the model-based and the measurement-based worlds exist.
In this talk, we will present our approach that aims to interchange and automate the extraction and transformation of workload specifications for measurement-based and model-based performance evaluation of session-based software systems. The approach comprises three main components: First, a domain specific language (DSL) allows the layered modeling of probabilistic workload specifications of session-based systems. Second, we automatically extract instances of this DSL from recorded session logs of production systems, e.g., employing Kieker. During this extraction process, different groups of customers with similar navigational patterns are identified using configurable clustering algorithms. Third, instances of the DSL are transformed to workload specifications for load generation tools and model-based performance evaluation tools. We will present a transformation to the common load testing tool Apache JMeter, including the Markov4JMeter extension developed in our previous work. Moreover, we will present our work in progress on transforming instances of the DSL to workload models in the Palladio Component Model. We evaluate our approach by comparing workload-specific characteristics (e.g., session lengths and arrival rates) from the recorded and the extracted/generated workload of the industry benchmark SPECjEnterprise2010.
|Document Type:||Conference or Workshop Item (Talk)|
|Keywords:||Workload specification, probabilistic workload, session-based software systems, load testing, performance prediction, Markov4JMeter, WESSBAS|
|Research affiliation:||Kiel University > Faculty of Engineering > Department of Computer Science > Software Engineering
Kiel University > Kiel Marine Science
|Date Deposited:||13 Dec 2014 20:13|
|Last Modified:||17 Dec 2014 23:14|
Actions (login required)