Article Preview
TopIntroduction
The implementation and delivery environment is complex and volatile for large-scale services, and there are various uncertainties in service performance, it results in poor quality of service, it can not meet the expected requirements or even fail, the reliability of service-oriented application execution is greatly affected (Xu L. D., 2011;Zheng Q. H. & Dong B.,2019), Building a stable service computing program becomes challenging. The service quality can mitigate service execution exceptions and risks based on performance prediction and analysis model (Blake M B & Cummings D J, 2012). Typical service performance analysis techniques mainly monitor service response time, service provider reputation and service reliability in real time. By collecting, measuring, predicting and analyzing service running status information, the problems can be found out in service execution in time (Dan A & Davis D, 2004; Emeakaroha V C & Netto M A S, 2012), service quality and performance are constantly improved and optimized in the cloud platform.
Research has yielded many achievements on service quality performance prediction and analysis models (Wei Y. & Blake M B.,2010). In order to find out the problems of quality of service in time, Leitner conducted dependency analysis on the service data execution logs, they found out the key services that affect the operation of the composite service, and the performance of the key services was monitored (Leitner P & Michlmayr A, 2010). In the process of service implementation, Rosenberg adopt the method of machine learning based regression (Rosenberg F & Leitner P, 2010), and the service quality prediction method was proposed based on autoregressive integrated moving average model (ARIMA) fitting (Hua Z. B. & Li M., Zhao J. F., 2013). By predicting the value of key service performance, service quality performance fluctuations are timely detected, thus the quality of service implementation is ensured.
However, with the explosive growth of publishing services in the cloud platform, the traditional service performance prediction model have been difficult to adapt to the large-scale features of massive service data based on full batch learning and its optimization algorithms (He Q. & Li N., 2014; Sun D. W. & Zhang G. Y., 2014; Lv Y. X. & Peng S. C., 2019). When dealing with large-scale service data, the service quality performance analysis model based on full batch forecasting needs to train a large amount of historical performance data, these will inevitably increase system extra cost and reduce operation efficiency. One of the fatal problems is difficult to meet the real-time computing requirements of large-scale services. Because the full batch learning method usually can only calculate the weight parameters of the predictive model function offline, and it is not suitable for real-time computing problems of large-scale services (Hashem I T & Yaqoob I, 2015; Moreno-Vozmediano R & Montero R S, 2013; Song L & Tekin C, 2016), how to effectively monitor the running quality of large-scale services and improve the performance evaluation methods, it needs further study.