A Reference Architecture for Online Performance Model Extraction in Virtualized Environments

Architectural Support for Software Performance in Continuous Software Engineering: A Systematic Mapping Study

“…A recurring issue is the need to integrate methods and tools into the continuous and DevOps pipelines. Hence, it is necessary to design and develop performance engineering approaches that can be integrated with other tools and methods used in the context of this study [SP29].…”

“…Software models are then aimed at checking either functional or non-functional properties. The analysis of software models in the context of incoming execution scenarios can suggest just-in-time refactoring/adaptation actions that keep the software behavior acceptable when these scenarios occur [8,9,10]. In the context of CSE processes, architectural models appear to have gained relevance, among others, for supporting performance-related decisions [11].…”

“…The three typical performance indices, namely, response time(40), utilization,(32) and throughput(17), are the most widely targeted ones in the considered papers. Although, on one hand, this can be seen as an expected outcome, on the other hand it is somehow surprising that memory(9) and network bandwidth(8) have been significantly less studied in this context. These two measures may play crucial roles in the performance assessment…”

Incremental Calibration of Architectural Performance Models with Parametric Dependencies

“…Walter et al [48] propose a tool to extract an architectural PM as well as performance annotation based on analysing monitoring traces. Similarly, other works [3,44,10] extract PM based on dynamic analysis. Krogmann et al [30,29] extract parametrized PCM based on hybrid analysis.…”

“…To address the high effort to create performance models when source code is available, researchers have suggested several approaches to extract PMs automatically. However, most of these approaches [10,48,3,33,44] have three shortcomings: First, they require instrumentation and execution of the whole system under study to extract the PM, which causes a high overhead and is not feasible at high frequency, e.g., after each source code commit. Second, they do not consider parametric dependencies.…”