Title: Auto-scale: automatic scaling of virtualised resources using neuro-fuzzy reinforcement learning approach

Authors: T. Veni; S. Mary Saira Bhanu

Addresses: Department of Computer Science and Engineering, National Institute of Technology, Tiruchirappalli 620015, Tamil Nadu, India ' Department of Computer Science and Engineering, National Institute of Technology, Tiruchirappalli 620015, Tamil Nadu, India

Abstract: Auto-scaling is an indispensable mechanism which facilitates automatic provisioning and de-provisioning of computing resources on-the-fly in accordance with fluctuating cloud service demands. The success of cloud computing necessitates that auto-scaling should improve both resource utilisation and end-user's quality-of-service. The dynamic and bursty workloads, interferences among virtual machines complicate the resource scaling process. In existing literature, threshold and reinforcement learning-based approaches are employed to enforce autoscaling policy. The threshold-based approach requires expertise knowledge of service-domains and reinforcement learning based approach suffers from the problem of 'curse-ofdimensionality'. In order to address these issues, a neuro-fuzzy reinforcement learning-based resource scaling approach is proposed to automatically adapt resource-scaling process to workload dynamics by considering both SLA constraints and resource utilisation. This approach requires no prior knowledge of service domains and is implemented on Xen-virtualised environment and tested with highly dynamic benchmark workload-RUBiS. Experimental results demonstrate that the proposed approach outperforms existing approaches.

Keywords: auto-scaling; cloud computing; reinforcement learning; virtualisation; neuro-fuzzy function approximation; workload dynamics; automatic scaling; virtualised resources; neural networks; fuzzy logic; cloud services.

DOI: 10.1504/IJBDI.2016.078400

International Journal of Big Data Intelligence, 2016 Vol.3 No.3, pp.145 - 153

Received: 21 May 2015
Accepted: 14 Jul 2015

Published online: 10 Aug 2016 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article