05 Fakultät Informatik, Elektrotechnik und Informationstechnik

Permanent URI for this collectionhttps://elib.uni-stuttgart.de/handle/11682/6

Browse

Filters

20191

Settings

Publication Type: search.filters.UBSTyp.Abschlussarbeit (Master)Author: search.filters.author.Hidiroglu, Alper

Search Results

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Context-aware load testing in continuous software engineering
    (2019) Hidiroglu, Alper
    To ensure adequate performance of a system, performance regressions have to be detected early in the software development process. Before a new software version is released, load tests should be applied on the system. In the context of continuous software engineering, it is crucial to keep delivery pipelines as short-running as possible in order to release software changes frequently. Since load tests typically take longer than functional tests, it is not possible to test for every possible workload scenario every time before a change is committed. It would be better to focus only on those load scenarios that are relevant for a given context, that consists, for example, of marketing campaigns, sports events, weather, etc. The goal of this work is to automatically generate load tests that test for the relevant load scenarios in the future. Thereby, we aim at reducing the resource usage and the test execution time. We develop a context description language to express contexts that can occur in the future. Our approach takes as input a context description and recorded request logs from the production system. It then uses the WESSBAS approach to calculate historical workload data from the recorded request logs. Based on the historical workload data and the passed context description, our approach then forecasts the future workload. The forecasted workload is processed and relevant load scenarios are identified that will occur in the future. Our approach then uses the WESSBAS approach again to automatically generate load tests that test for the identified load scenarios. We evaluated our approach with a real-world data set, that contains recorded requests from the Student Information System (SIS) of the Charles University in Prague. The evaluation shows that contexts help to reduce the testing effort and to focus only on the relevant workload scenarios. However, the evaluation also shows that our approach has limitations regarding the accuracy of the forecasted workload. Load tests, that are generated from inaccurately forecasted workload, do not test for the relevant load scenarios.