NFC-ARP: neuro-fuzzy controller for adaptive resource provisioning in virtualized environments

Abstract

In virtualized environments like cloud, elasticity is an indispensable feature which facilitates on-demand provisioning or de-provisioning of resources in order to meet the service demands of end-users. The key factors such as dynamic and bursty workloads, shared resource contention, and complex virtualized infrastructures present significant challenges in on-demand elastic resource provisioning process. Fuzzy control system is a prominent mechanism for efficient resource provisioning in virtualized environments. However, in existing fuzzy controllers, the design parameters such as rule base and membership functions are determined manually on a trial-and-error basis that are not adaptive to time-varying workloads. This results in performance degradation of applications hosted on the virtual machines. In order to address this issue, in this paper, a neuro-fuzzy controller-based adaptive resource provisioning (NFC-ARP) approach is extended to reconfigure the various resources of virtual machines in response to workload variations. The advantage of NFC-ARP is that it exploits the strength of both fuzzy logic and neural networks, thereby adjusting fuzzy controller’s design parameters based on workload variations. NFC-ARP is implemented on Xen virtualized environment, and the experimental results demonstrate that NFC-ARP outperforms the existing state-of-the-art controllers in terms of minimal resource utilization cost and SLA violations.