Search

A hybrid of supervised (artificial neural network), unsupervised (clustering) machine learning, and soft computing (interval type 2 fuzzy logic system)-based load balancing algorithm, i.e., clustering-based multiple objective dynamic load balancing technique (CMODLB), is introduced to balance the cloud load in the present work. Initially, our previously introduced artificial neural network-based dynamic load balancing (ANN-LB) technique is implemented to cluster the virtual machines (VMs) into underloaded and overloaded VMs using Bayesian optimization-based enhanced K-means (BOEK-means) algorithm. In the second stage, the user tasks are scheduled for underloading VMs to improve load balance and resource utilization. Scheduling of tasks is supported by multi-objective-based technique of order preference by similarity to ideal solution with particle swarm optimization (TOPSIS-PSO) algorithm using different cloud criteria. To realize load balancing among PMs, the VM manager makes decisions for VM migration. VM migration decision is done based on the suitable conditions, if a PM is overloaded, and if another PM is minimum loaded. The former condition balances load, while the latter condition minimizes energy consumption in PMs. VM migration is achieved through interval type 2 fuzzy logic system (IT2FS) whose decisions are based on multiple significant parameters. Experimental results show that the CMODLB method takes 31.067% and 71.6% less completion time than TaPRA and BSO, respectively. It has maintained 65.54% and 68.26% less MakeSpan than MaxMin and R.R algorithms, respectively. The proposed method has achieved around 75% of resource utilization, which is highest compared to DHCI and CESCC. The use of novel and innovative hybridization of machine learning, multi-objective, and soft computing methods in the proposed algorithm offers optimum scheduling and migration processes to balance PMs and VMs. [ABSTRACT FROM AUTHOR]

Cloud computing is an emerging paradigm that offers various services for both users and enterprisers. Scheduling of user tasks among data centers, host and virtual machines (VMs) becomes challenging issues in cloud due to involvement of vast number of users. To address such issues, a new multi-criteria approach i.e., technique of order precedence by similarity to ideal solution (TOPSIS) algorithm is introduced to perform task scheduling in cloud systems. The task scheduling is performed in two phases. In first phase, TOPSIS algorithm is applied to obtain the relative closeness of tasks with respect to selected scheduling criteria (i.e., execution time, transmission time and cost). In second phase the particle swarm optimization (PSO) begins with computing relative closeness of the given three criteria for all tasks in all VMs. A weighted sum of execution time, transmission time and cost used as an objective function by TOPSIS to solve the problem of multi-objective task scheduling in cloud environment. The simulation work has been done in CloudSim. The performance of proposed work has been compared with PSO, dynamic PSO (DPSO), ABC, IABC and FUGE algorithms on the basis of MakeSpan, transmission time, cost and resource utilization. Experimental results show approximate 75% improvement on average utilization of resources than PSO. Processing cost of TOPSIS–PSO reduced at approximate 23.93% and 55.49% than IABC and ABC respectively. The analysis also shows that TOPSIS–PSO algorithm reduces 3.1, 29.1 and 14.4% MakeSpan than FUGE, ant colony optimization (ACO) and multiple ACO respectively. Plotted graphs and calculated values show that the proposed work is very innovative and effective for task scheduling. This TOPSIS method to calculate relative closeness for PSO has been remarkable. [ABSTRACT FROM AUTHOR]

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. • The work introduced a new hybridized soft computing inspired technique named Clustering based Artificial Neural Network for Dynamic Load Balancing (CANN-DLB). • Back-propagation ANN and K-Means algorithm are performed to calculate the optimized VM loadin cloud systems for improving QoS. • Runtime incoming user tasks are allocated to underloaded VMs to maintain load balancing using PSO technique. • CANN-DLB achieved 0.99% of resource utilization which is higher than existing algorithms and 69.5%, 96.9%, 96.0% and 97.4% less processing cost than previous algorithms respectively. • For Proposed CANN-DLB, degree of imbalance is reduced to 87.12%, 87.12%, 87.57% and 81.78% than CM-eFCFS, Static RR, MinMin and MaxMin method respectively. [ABSTRACT FROM AUTHOR]