Novel hybrid ANN and clustering inspired load balancing algorithm in cloud environment

Abstract

Load balancing in the cloud environment is a NP-hard problem. To address this problem, a new hybridized soft computing inspired technique named Clustering based Artificial Neural Network for Dynamic Load Balancing (CANN-DLB) is introduced. Artificial Neural Network aims to achieve an optimized load of VMs by training the cloud environment. Back-propagation ANN is performed to calculate the optimized Virtual Machine (VM) load in cloud systems for improving QoS. The CANN-DLB technique uses K-means clustering algorithm over calculated VM loads and clusters them into under-loaded and over-loaded VMs. The scheduling of homogeneous independent non-preemptive tasks is performed using Particle Swarm Optimization (PSO) technique. CloudSim (Cloud Simulator) tool has been used to implement the proposed algorithm. The performance of the CANN-DLB technique has been achieved for Space Shared and Time Shared VM task scheduler methods. The simulation results of the CANN-DLB algorithm are compared with existing load balancing and scheduling algorithms with the objectives to improve load balance fairness. Results of CANN-DLB have achieved 69.5%, 96.9%, 96.0% and 97.4% less processing cost and 87.12%, 87.12%, 87.57% and 81.78% less degree of imbalance compared to CM-eFCFS, Static RR, MinMin and MaxMin method respectively. The system fairness with respect to system load is calculated to analyze CANN-DLB algorithm performance on heavy load. The proposed algorithm achieved 12% higher load fairness than the existing Two-level method and proves that the introduced model works better in dynamic scenarios. Plotted graphs show that the proposed idea is innovative for load balancing in a dynamic cloud environment. This hybridization of the ANN and K-means clustering method produces remarkable results as compared to existing algorithms for different cases.