An Automatic Approach to Detect Software Anomalies in Cloud Computing Using Pragmatic Bayes Approach

Abstract

Software detection of anomalies is a vital element of operations in data centers and service clouds. Statistical Process Control (SPC) cloud charts sense routine anomalies and their root causes are identified based on the differential profiling strategy. By automating the tasks, most of the manual overhead incurred in detecting the software anomalies and the analysis time are reduced to a larger extent but detailed analysis of profiling data are not performed in most of the cases. On the other hand, the cloud scheduler judges both the requirements of the user and the available infrastructure to equivalent their requirements. OpenStack prototype works on cloud trust management which provides the scheduler but complexity occurs when hosting the cloud system. At the same time, Trusted Computing Base (TCB) of a computing node does not achieve the scalability measure. This unique paradigm brings about many software anomalies, which have not been well studied. This work, a Pragmatic Bayes approach studies the problem of detecting software anomalies and ensures scalability by comparing information at the current time to historical data. In particular, PB approach uses the two component Gaussian mixture to deviations at current time in cloud environment. The introduction of Gaussian mixture in PB approach achieves higher scalability measure which involves supervising massive number of cells and fast enough to be potentially useful in many streaming scenarios. Wherein previous works has been ensured for scheduling often lacks of scalability, this paper shows the superiority of the method using a Bayes per section error rate procedure through simulation, and provides the detailed analysis of profiling data in the marginal distributions using the Amazon EC2 dataset. Extensive performance analysis shows that the PB approach is highly efficient in terms of runtime, scalability, software anomaly detection ratio, CPU utilization, density rate, and computational complexity.