Detection of Performance Anomalies in Cloud Services: A Correlation Analysis Approach

Abstract

The prevalence of ubiquitous computing and communication has coined the term of cloud computing through which, software, infrastructure and platform can be provided as a service. Software as a service (SaaS) is getting an increasing potential as a cloud-based option for using software applications in a payper-use manner. A critical challenge in SaaS model is continuous attestation of the compliance with quality of service (QoS) metrics stated in SLAs. In this paper, we propose a method for detecting performance anomalies in cloud software services. The proposed method uses correlation analysis between computing resources utilization and workload characteristics. This is done by comparing the correlation values to a reference load test values performed before the SaaS deployment to identify deviations and notify the system administrator about it. The testing scenario operates in two steps. First, running a standard benchmark on a virtual machine to simulate workload and record the correlation between workload and available computing resources utilization (i.e., CPU, RAM, HDD, and Network). Second, the same benchmark is executed again but with changing the workload characteristics through injecting additional queries or changing the computing resources configuration values of the virtual machine. The changes are only present on specific time points to testify the detection rate. Results on standard benchmarks TPC-C, TPC-D and TPC-W showed a promising detection rate that can assure SLA targeted quality aspects such as reliability, scalability and security.