Load Balancing in Cloud Computing

Abstract

Cloud Computing is a technology that focuses on the designing of the computing system, develops applications and uses the existing system to build software. Cloud Computing is designed to provide computing and other services like water, gas, telephone bill, electricity, etc. Users can access these services as per their needs and they get charged for the amount of time they have used those services. This model of charging is called pay-per-usage of services. newline newlineCloud Computing provides several advantages like reducing the cost of IT services, flexibility, and scalability. Though highly advantageous for many services, Cloud Computing usage has its limitations too like security, privacy, and response time, etc. Response time is one of the issues related to improper load balancing in the cloud which hinders the efficiency of the cloud service provider in catering the customers requirements efficiently. Most of the reviewed algorithms like Min-Min, Max-Min, and Ant Colony Optimization have an issues with response time while balancing the incoming load on to the cloud server. newline newlineThe objective of this research work is to design an algorithm for efficient load balancing system on Cloud Computing. A detailed analysis of response time, throughput and overheads related to Virtual Machines (VM s) used for providing cloud services have been carried out in this thesis where VM s act as a backbone of Cloud Computing. newline newlineThe whole thesis comprises of seven chapters. Chapter one gives insight into the evolution of Cloud Computing, its service and deployment model together with its characteristics. The strength of cloud usage and its demand forecast given by Gartner in India have also been discussed here. newline newlineChapter two focuses on the load balancing issue, VM migration process with its limitations and the concepts of Virtualization and VM. Load Balancing algorithms are also discussed in their respective categories (like Static and Dynamic Load Balancing). Simulation tools used for the simulation purpose are discussed at the end of the chapter. newline newlineChapter three comprises of literature review covering different design issues related to good Load Balancing algorithms. In the same chapter, research objectives are also newline newlineoutlined. Most of the research objectives are directed for the improvement of response time and efficient utilization of available resources. newline newlineThe first research objective is detailed in chapter four and an analysis has been done of the load imbalance in the distributed file system. For load rebalancing, the whole VM or chunk server in-place of chunks of loads have been migrated to the different server. Division of the load or incoming tasks into smaller sub-tasks aimed for the migration of those tasks, resulted in increased response time. It has been concluded that load rebalancing can reduce the load migration time or response time or the load movement cost. newline newlineChapter five proposes a Load Balancing approach called Priority Based Load Balancing and the comparison of the results with other algorithms. Here customers have been prioritized on the basis of their Quality of Service demand and geographical location. It is on the basis of resources (or the number of VM s) that the Data Centers (DC) have also been prioritized. Priority is assigned to each customer and DC and a mapping between priority customer and prioritized DC is accomplished. Using this approach, appropriate DCs from the same geographical location are assigned to the appropriate customer, which steered an indirect reduction in the response time. newline newlineChapter six gives insight into another load balancing approach called Hybrid approach for proper utilization of resources. The advantages of Equally Spread Current Execution (ESCE) algorithm and Double Priority algorithm have been earmarked to propose a Hybrid algorithm. ESCE is advantageous in the ensuring the equal distribution of the incoming load on the available resources of DC and Double Priority algorithm is advantageous in assigning priority to the incoming tasks and the VM s present in DC. It is for the efficient utilization of resources that the mapping of the tasks and VMs is executed. The tasks with the least size are served by the VM with least MIPS capacity. It is with the help of this hybrid algorithm that the starvation problem (insufficiency of resources) of the some large size tasks is also addressed. newline newlineThe future scope of the research work has been discussed in the seventh chapter. Since cloud is an emerging and evolving technology, a lot of research can be done in the areas of green and utility cloud, along with some other issues like security and vendor lock-in in the near future. newline newline newline