An adaptive threshold policy for host overload detection in cloud data centre

Excessive resource usage in a cloud computing system causes an increase in operational costs. In contrast, shortage of resources results in increased load on the server leading to Service Level Agreement Violation (SLAV) and reduced Quality of Service (QoS). Since, the workload is highly dynamic in nature, maintaining the utilization at optimal levels to effectively consume energy while keeping SLA integrated is a challenging task. To deal with the variability of the workload, the current research study focuses on calculating the upper threshold using past Central Processing Unit (CPU) utilization with the help of an adaptive threshold policy. The proposed Pn estimator based Adaptive Threshold policy (P_n_AT_H P) scheme implements an estimator Pn that periodically analyses the previous CPU utilization data of a host machine and sets the upper threshold accordingly. The results so obtained against traditional schemes of overload detection, show improvement in the performance metrics. According to the simulation analysis, the proposed P_n_AT_H P host overload detection scheme shows an improvement of 41.83% and 44.70%, compared to existing LRMMT and SNMMT schemes, in terms of the combined metric of energy, SLAVs, and the number of Virtual Machine (VM) migrations.