Font Size: a A A

Research On Cluster Workloads And Scheduling Policies In Cloud Computing

Posted on:2015-03-28Degree:MasterType:Thesis
Country:ChinaCandidate:J YingFull Text:PDF
GTID:2268330428964517Subject:Computing applications technology
Abstract/Summary:PDF Full Text Request
More and more users use cloud computing services, the cluster in cloud data center is themain cloud service provider. Due to the complexity of the cloud environment, understandingthe characteristics of cloud workloads is the key to improving the system performance andoptimizing the cloud cluster. However, workload characterization of cloud, especially in alarge scale production environment, has not been fully studied yet. Virtualization technologyis the core technique in cloud environment and dynamic scheduling of virtual machines(VMs)is an effective way to improve the cloud data center resource utilization, however, there is nostandard algorithm in this field. This thesis conducts a comprehensive analysis of the cloudand studies the VM scheduling algorithm based on the workload characteristics, alsoconsidering reducing energy and ensuring customer’s service level agreements(SLA).This thesis firstly investigates the characteristics and shortcomings of the traditional datacenter and studies the characteristics of the cloud data center. Then, workload monitoringtechnology in the cloud cluster and VM workload gathering techniques based on libvirt aresturdied. The thesis also introduces the concept and characteristics of scheduling in the cloudand a detailed analysis of the scheduling mechanism in cloud environment based onOpenStack is provided. Then we analyze the cloud simulation techniques based on CloudSim.Subsequently, to gain insight on cloud data center workloads, this thesis collects aone-week workload trace from a public cloud provider, including1082VM instances and100physical nodes, logged from April11to April17,2013. Through characterizing the workloadof cloud data center from different angles, including VM distribution and schedulingmechanism, distribution of VM memory, VM CPU and I/O characteristics, physical nodeworkload characteristics, relationship between VM workload and the physical node workload,this thesis concludes a set of observations from the workload characterization and usesanalysis to derive several implications for performance optimization solutions.Finally, on the basis of the cloud cluster workload characterization, this thesis studies thecharacteristics and the model of VM scheduling in the cloud, then proposes a optimizedalgorithm called ERSG(Energy Reducing and SLA Guarantee). The algorithm optimizes thethree steps of the VM scheduling, including scheduling opportunity for the VM, basing onthe workload forecasting model; selection strategy for the VM, ensuring the users’ SLA; thechoice of the target node, probabilistic analysis method introduced. Through the simulationexperiments which simulating a cloud data center, results show that ERSG algorithm is more excellent than other algorithms in energy reducing and SLA guarantee.
Keywords/Search Tags:cloud data center, workload characterization, VM scheduling, energy reducing, SLA
PDF Full Text Request
Related items