عنوان انگلیسی مقاله:
Performance evaluation of web service response time probability distribution models for business process cycle time simulation
ترجمه فارسی عنوان مقاله:
ارزیابی عملکرد مدلهای توزیع احتمال پاسخ زمان سرویس وب برای شبیه سازی چرخه فرآیند کسب و کار
Sciencedirect - Elsevier - The Journal of Systems & Software, 161 (2020) 110480: doi:10:1016/j:jss:2019:110480
Context: The adoption of Business Process Management (BPM) is enabling companies to improve the pace
of building new capabilities, enhancing existing ones, and measuring process performance to identify bottlenecks. It is essential to compute the cycle time of the process to assess the performance of a business
process. The cycle time typically forms part of service level agreements (SLAs) and is a crucial contributor
to the overall user experience and productivity. The simulation technique is versatile and has broad applicability for determining realistic cycle time using historical data of web service response time. BPM tools
offer inadequate support for modeling input data used in simulation in the form of descriptive statistics or standard probability distributions like normal, lognormal, which results in inaccurate simulation
Objective: We evaluate the effectiveness of different parametric and non-parametric probability distributions for modeling data of web service response time. We further assess how the choice of probability
distribution impacts the accuracy of the simulated cycle time of a business process. The work is the first
of such a study using real-world data for encouraging Business Process Simulation Specification (BPSim)
standard setters and BPM tools to enhance their support for such distributions in their simulation engine.
Method: We consider several parametric and non-parametric distributions and explore how well these
distributions fit web service response time from extensive public and a real-world dataset. The cycle time
of the business process of a real-world system is simulated using the identified distributions to model the
underlying web service data.
Results: Our results show that kernel distribution is the most suitable choice, followed by Burr. Kernel
outperforms Burr by 86.63% for the public and 84.21% for the real-world dataset. The choice of distribution affects the percentile ranks like 90 and above than the median. The use of single-point values
underestimates cycle time values at higher percentiles.
Conclusion: Based on our empirical results, we recommend the addition of kernel and Burr to the current
list of distributions supported by BPSim and BPM tools.
Keywords: Simulation input modeling | Parametric distributions | Non-parametric distributions | Performance evaluation | Web service response time | Cycle time