Runtime reliability fractional distribution change analytics against cloud-based systems DDoS attacks

With the help of the Software as a Service (SaaS) delivery model, the rapid advancement of cloud computing has become the most prevalent distributed computing paradigm. A large number of application vendors and developers choose to integrate cloud-hosted Application Program Interfaces (APIs) into their systems as system components to construct new and value-added cloud-based systems. When executed in an open cloud environment, each system component is constantly at risk of Distributed Denial of Service (DDoS) attacks. Such cloud-based systems are challenged by reliability fluctuations when a system component is attacked. A change in the reliability of the remote system components, e.g., performance decline or runtime anomalies, can threaten the stability of the entire cloud-based system. To enable timely reliability assurance against cloud-based systems DDoS attacks, it is necessary to analyze runtime reliability of its system components. In this paper, we formally present a new model for evaluating the reliability of the system components based on concept drift. Based on the model, we propose a novel method named runtime reliability anomaly detection (RAD), leveraging the Singular Value Decomposition (SVD) technique. RAD analyzes the reliability of a system component during its operation by detecting peaks in Fractional Distribution Change (FDC) within its reliability time series data. Specifically, it calculates the Jensen Shannon divergence between historical and up-to-date reliability data streams, based on feature vectors that are dimensionality-reduced using SVD. The results of extensive experiments conducted on two public cloud APIs performance datasets demonstrate the effectiveness and efficiency of RAD.