Non-linear analysis of bursty workloads using dual metrics for better cloud resource management

The assumption that enterprise workloads are steady-state could make their resource provisioning ineffective. The current study aims to address this challenge by performing a non-linear analysis on a set of synthetic bursty workloads. The research involves building resource-provisioning models using non-linear metrics, namely hurst exponent and sample entropy. Performance of the proposed approach was compared with baseline reactive approach and the index of dispersion approach using the NASA dataset. The proposed approach had a sensitivity of 70% and specificity of 90%. The reactive approach had a sensitivity and specificity of 55% and 84%, respectively while the index of dispersion had a sensitivity of 61% and specificity of 82%. The current study also displayed a 71% reduction in error-count compared to the baseline reactive approach.