PSO BASED SCHEDULING TECHNIQUES TO IMPROVE QoS PARAMETERS IN CLOUD COMPUTING

Abstract

Cloud computing provides the services either in the form of software application or hardware infrastructure on the basis of pay per use over the internet. There are lots of challenges in the field of cloud computing due to improper management of cloud resources. Scheduling challenges occurs due to dispersion, uncertainty and heterogeneity of resources that are not resolved with traditional resource management mechanisms. Over provisioning and under provisioning types of problem occurs in cloud environment due to which cloud resources are not utilizing properly. One of the challenging features of cloud computing is to provide on demand huge number of resources to the user as per their need (elasticity and scalability) and satisfy the quality of service (QoS) parameters like reliability, elasticity, deadline, priority of task etc. to minimizing the makespan time and task rejection ratio. Therefore we need an efficient scheduling algorithm that utilize the cloud resources properly and improve the QoS parameters. Following are the major objectives of the thesis: 1. To develop a load balancing algorithm with elasticity concept that balances the workload among the virtual machine and analyzes the QoS parameters (execution time, makespan time, task rejection ratio) considering deadline as constraint. 2. To propose a dynamic transfer function based modified binary PSO to solve the real world discrete problems and analyze the QoS parameters in cloud computing 3. To Develop a PSO based multi-objective scheduling algorithm to analyze the effect of execution time, execution cost and energy consumption in the field of cloud computing. Objective 1: Development a load balancing algorithm with elasticity Most of the published scheduling algorithm deals with only one parameter either scheduling the upcoming tasks to optimize the required parameter or scalability within user defined deadline constraint. It is difficult to predict and calculate all possible task-resource mapping in cloud environment. One of the important aspects of scheduling is to balance the workload among the cloud resources (virtual machines) and monitors the load at each virtual machine continuously. Contribution: To solve the issue of load balancing with elasticity in cloud computing, we have developed a II dynamic scheduling algorithm that balances the workload among all the virtual machines with elastic resource provisioning and deprovisioning based on the last optimal k-interval considering deadline as constraint. Developed algorithm distributes the tasks and adds the cloud resource if task rejection ratio is more than the service level agreement (SLA) defined threshold value. The computational results proved that the developed algorithm decreases the makespan time (up to 6% in comparison with min-min, up to 15% from shortest job first and up to 20 % from first come first serve algorithm) and task meet with deadline ratio of developed algorithm is up to 93% compare to other algorithm (min-min up to 80.8 %, shortest-job-first up to 75% and first come first serve 73%) in all conditions. Objective 2: Development of a dynamic transfer function based modified BPSO Binary particle swarm optimization (BPSO) is used to solve discrete optimization problems but it does not maintain the good balance between exploration and exploitation for transfer function. Contribution: To overcome this problem, we have developed a dynamic transfer function (TFP-BPSO) for BPSO which provides better exploration at the early stage by high flipping of bit of particle position for any velocity and it has the ability to move from exploration to exploitation in the intermediate stage of execution. It provides the stronger exploitation (less probability of flipping of bits) in the last stage of execution. Results proved that developed TFP-BPSO algorithm reduces the execution time (up to 10% compare with BPSO and up to 20% compare with FCFS) makespan time (up to 15% comparison with BPSO and up to 32% comparison with FCFS). Throughput has been increased (up to 20% compare with BPSO and up to 40% compare with FCFS) in better way than existing algorithm like first come first serve and BPSO. Objective 3: PSO based cost and energy efficient scheduling algorithm with deadline constraint The main purpose of cloud service provider is to maximize the profit and minimize the energy consumption from cloud infrastructure while cloud users want to execute their applications in minimum time and execution cost. There is always a conflict between execution cost and time parameters because low cost resources are less computation oriented than expensive. So a trade-off solution is required to optimize both the parameters at the same time. The rapid growths in the demand of computational power tends to massive growth in cloud data centers III and require large amount of energy consumption in cloud data centers which has become a serious threat to the environment. To reduce the energy consumption in cloud computing is a challenging problem due to incompatibility between workstation (physical machine) and unpredictable users demand. Contribution: We have proposed a resource allocation model for processing the applications efficiently and particle swarm optimization based scheduling algorithm that not only optimize execution cost and time but also reduce the energy consumption of cloud data centers considering deadline as constraint. The developed algorithm has been simulated at cloudsim and it is observed that it reduces the execution time (up to 8% from existing PSO, 15% honey bee, 20 % min-min algorithms), makespan time (up to 10% from existing PSO, 20% honey bee, 18.8 % min-min algorithms) execution cost (up to 8% from existing PSO, 12% honey bee, 15 % min-min algorithms), task rejection ratio (up to 8% from existing PSO, 23% honey bee, 19.6 % min-min algorithms), energy consumption (up to 7% from existing PSO, 11% honey bee, 18 % min-min algorithms) and increase the throughput (up to 6.5% from existing PSO, 9.8% honey bee, 12.6 % min-min algorithms) in comparison to PSO, honey bee and min-min algorithm. The thesis is organized into six chapters including the introduction. The thesis is organized into six chapters. CHAPTER 1: INTRODUCTION In the introduction part, details with general concepts of cloud services, motivation, research challenges, objectives and contributions of this thesis. CHAPTER 2: FUNDAMENTALS AND SCHEDULING TECHNIQUES In this chapter, we discuss the existing resource provisioning techniques, advantages of resource provisioning in the field of cloud computing, static and dynamic scheduling algorithm, classification of scheduling algorithms in terms of heuristic, meta-heuristic and hybrid algorithm. Further resource allocation model and simulation tool are discussed in brief that are used to measure the performance of the scheduling algorithm. IV CHAPTER 3: LOAD BALANCING WITH ELASTICITY USING HEURISTIC TECHNIQUE The details about the development of a cloud resource broker architecture and dynamic scheduling algorithm that is able to automatically manage and monitor the virtual machines to minimize the QoS parameters based on the last optimal k-interval of considering deadline as constraint and simultaneously fulfill the objective of elasticity in cloud environment have been discussed. CHAPTER 4: DYNAMIC TRANSFER FUNCTION BASED MODIFIED BINARY PSO FOR SCHEDULING THE TASKS Development of dynamic transfer function (TFP-BPSO) based BPSO algorithm that provides better exploration at the early stage, its move from exploration to exploitation in the intermediate stage of execution and provides the stronger exploitation in the last stage of execution to improve QoS parameters have been discussed. CHAPTER 5: MULTI-OBJECTIVE SCHEDULING ALGORITHM USING PSO Particle swarm optimization (PSO) based scheduling algorithm and resource allocation model for improvement of QoS parameters have been discussed. CHAPTER 6- CONCLUSIONS AND FUTURE WORK This chapter concludes the work reported in the thesis and discusses about future research directions.