Performance Difference Prediction in Cloud Services for SLA-Based Auditing

Cloud computing allows individuals and organizations outsource their applications to cloud due to the flexibility and cost savings. However, cloud service providers (CSPs) may offer reduced resources to tenant by virtualization technology for illegal economic benefit. Most cloud tenants don´t have specialized knowledge to detect the CSPs´ service level agreement (SLA) disruption. To address this issue, we propose a worst-case performance prediction scheme to audit the SLA disruption of cloud by comparison between actual performance cost and the predicted worst-case performance cost. Firstly, we insert labels into application, and collect the running time information on cloud on a small input dataset. The historical information is used to identify the frequent block. Then, we construct the performance cost function for each hotspot block, and resort to curve fitting method to obtain the performance cost function of each frequent block. Finally, we get the worst-case performance prediction by performance estimation model and performance cost function of each one on large input dataset. The experiments show that our proposed scheme can achieve high sensitiveness on cloud SLA and fulfill cloud SLA auditing.