Distribution of virtual machines using static and dynamic load balancing in cloud computing / Misbah Liaqat

Misbah , Liaqat (2017) Distribution of virtual machines using static and dynamic load balancing in cloud computing / Misbah Liaqat. PhD thesis, University of Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (159Kb)
Preview	PDF (Thesis PhD) Download (2846Kb) | Preview

Abstract

Cloud computing, a user-centric computational model, is flexible paradigm of deploying and sharing distributed services and resources with the pay-per-use model. With virtual machine (VM) technology and data centers (DCs), computational resources, such as memory, central processing unit (CPU), and storage, are dynamically reassembled and partitioned to meet the specific requirements of end users. The demand’s growth for cloud services is presenting considerable challenges for cloud providers to meet the requirements and satisfaction of end users. Virtualization technology reduces cloud operational cost by increasing cloud resource utilization level. In addition, the ever growing computational demands of users call for e_cient cloud resource management to avoid SLA violation. Virtualization co-locates multiple virtual machines (VM) on a single physical server to share the underlying resources for e_cient resource management. However, the decision about ”what” and ”where” to place workloads significantly impacts performance of hosted workloads. Load balancing between physical servers is important to avoid dangerous hot spots in the Cloud; in fact, overload situations are dangerous since they can easily lead to resource shortage and, at the same time, they can a_ect hardware lifetime, thus undermining data center reliability. Existing cloud schedulers consider a single resource (RAM) to co-locate workloads that as a result lead to SLA violation due to nonoptimal VM placement. In addition, allocation of VMs based on traditional scheduler ine_ciently balance the workload distribution that leads to extended the application execution time. Furthermore, exiting studies incorporates the migration technique in order to balance the load after the initial placement of workload, which leads to the maximum numbers of migrations. Therefore, to overcome these issues, this study propose the efficient load balancing solutions to uniformly distribute the workload among the physical servers. The initial VM placement method called Static Multi Resource based Sched uler (SMRS), is designed to enhance the application execution time while balancing the CPU utilization without VM migrations. In addition, the Dynamic Multi Resource based Scheduler (DMRS) method is proposed to minimize the number of migrations after the initial placement of workload. We performed the real time experiments using the Open- Stack cloud to highlight the e_ciency of SMRS and DMRS solutions. Moreover, this study proposed the mathematical model for SMRS and DMRS method. To validate the correctness of the mathematical model, the empirical results and mathematical results are compared based on the CPU utilization, application execution time, and numbers of VM migrations as a performance metrics. The e_ectiveness of the proposed solution is evaluated by comparing their empirical results with well-known standard OpenStack nova scheduler. Experimentally, we have shown that our proposed method has lessened application execution time by 50% when compared with standard OpenStack cloud in static environment. In dynamic environment, the performance gain is reported up to 85% and 94.4% based on application execution time and CPU utilization. The improvement in application execution time increases the usability of cloud data centers.

Actions (For repository staff only : Login required)