VTB-RTRRP: Variable Threshold Based Response Time Reliability Real-Time Prediction

To provide high quality of network service, the response time reliability (RTR), the probability that the response time of a service is within the maximum allowable time determined by users, attracts researchers from both academic and industry. For those networked systems that can adjust their configuration adaptively, such as cloud computing systems, ad hoc networks, and network function virtualization systems, it is essential to predict the RTR of the system, and use the prediction result as a constraint to optimize the system configuration, e.g., tradeoff between reliability and energy efficiency. In general, the RTR changes along with the workload of the system. Due to the complex communication process, it is not easy to derive the analytical model between workload and RTR, and the maximum response time that users can tolerate also changes with the system workload according to users' experience. This paper proposes a variable thresholdbased RTR real-time prediction framework based on data-driven approaches to forecast the RTR for networked systems. In this framework, the response time threshold varies with the workload, reflecting both users' acceptable and perceived response time, and a real-time updating model between workload and RTR is built for RTR prediction based on the latest monitored dada. Finally, a communication network is taken as the example to validate the effectiveness of the proposed method from different perspectives. The experimental results illustrate that this continually updating model can obtain more accurate RTR prediction.