Performance Curve Profiling and Gated Recurrent Unit Based State Detection for Cloud Native Microservices

Performance Curve Profiling and Gated Recurrent Unit Based State Detection for Cloud Native Microservices

연구 분야: Software Development

학회: International Conference on Service Science

Cloud-native technologies have been widely adopted by practitioners and companies with various needs to ensure quality of service by deploying cloud-native applications in cloud servers and using resource auto-scaling techniques to meet changing demands. Although resource auto-scaling techniques in cloud platforms usually maintain the stability of microservices, sometimes they still fail. In this paper, a hybrid performance curve profiling and gated recurrent unit based state detection method (PGS) is developed for identifying the states for cloud-native microservices and provide a basis for resource auto-scaling. Because the microservice states arenot able to identified from transient performance indicators, a performance curve profiling method (PCP) is proposed in PGS to assist identifying microservice states from a global perspective. For constantly changing microservice states, a gated recurrent unit based state detection method (GSD) is proposed to improve the accuracy of states recognition. GSD divides microservice states into three categories based on the slope of the performance curve. The experimental results show that PGS achieves an accuracy of 96.09% for microservice state detection by PCP assisting GSD to judge from a global perspective, which is higher than other methods.