Font Size: a A A

Research On Network Resources Management And Scheduling Technologies For Data Centers

Posted on:2016-02-01Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y Y LiFull Text:PDF
GTID:1108330482457706Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
A great change of cloud computing is to provide elastic and scalable ser-vices which take from the pool of hardware resources by means of virtualization technology. The essence is to share infrastructures such as data centers to mul-tiplex the resources. It is an urgent need to address the sharing problem of data center network resources. In order to enhance the performance of applications, all tenants will try to compete for bandwidth, which affects the performance of other applications. In addition, it has been shown that various data schedul-ing schemes have different influences on the operation cost of data centers. Hence, it is important to manage and schedule data center network resources to guarantee the quality of services, improve the network resource utilization and optimize the operation cost of data centers through service isolation, band-width allocation and request dispatching. This dissertation studies the network resources management and scheduling technologies for data centers from the perspective of intra-datacenter, inter-datacenter and user-to-datacenter respec-tively. The contribution can be summarised in the following.(1) By means of virtualization, computing and storage resources are ef-fectively multiplexed by different applications in cloud data centers. However, there lack useful approaches to share the network resources of data centers. Invalid network sharing not only degrade the performance of applications, but also affect the efficiency of data center operation. To guarantee network perfor-mance of applications and provide fine-grained service differentiation, in this paper, we propose a differentiated bandwidth guarantee scheme for data cen-ter networks. Utility functions are constructed according to the throughput and delay sensitive characteristics of diverse applications. Seeking to maximize the utility of all applications, the problem is formulated as a multi-objective optimization problem. We address this problem using a heuristic algorithm: NSGA-II.(2) Cloud service providers are building out geographically distributed networks of data centers around the world. It is customary for cloud service providers to distribute their data replicas at multiple geographic locations to mitigate user latency and increase service availability. In this paper, we treat the content distributed from one data center to multiple data centers as a mul-ticast session. We study the problem of maximizing the capacity utilization of inter-datacenter networks while maintaining fairness among multiple mul-ticast sessions. A bandwidth allocation algorithm based on max-min fairness is developed, named Blossom. Blossom leverages a fully polynomial time ap-proximation scheme to accelerate the bandwidth allocation, while achieving an approximation that is (l-ε)-optimal.(3) Many inter-datacenter bulk transfer schemes are suggested to improve the bandwidth utilization, reduce cost on inter-datacenter traffic and decrease inter-domain traffic respectively. However, in practice, administrators of data centers not only pay attention to the bandwidth utilization, but also consider the link cost, quality of services and so on. A multi-attribute decision making based method is therefore proposed to schedule inter-datacenter bulk data. The multi-attribute aware scheduling problem is modeled on a time expanded graph and formulated as a minimal cost flow problem. Simulation results show that this strategy can take multiple attributes into consideration, and can find the bulk transfer route with the best integrated evaluation.(4) Nowadays, energy cost has become increasingly important fraction of data center operation cost. Researcher’s attentions are attributed to propose many methods to reduce the electricity cost for data centers. However, another significant operation cost, Internet bandwidth cost, has been neglected. Because both the electricity prices and bandwidth prices are varied with time and regions. It is an opportunity to reduce data center bandwidth cost by dispatching user requests to suitable data centers. But it is difficult to minimize the operation cost while response to user requests timely. As a solution, we propose an online control policy based on Lyapunov optimization framework to achieve close-to-optimal performance with a tradeoff between cost and delay.
Keywords/Search Tags:data center network, bandwidth allocation, bulk transfer, operation cost optimization, request scheduling
PDF Full Text Request
Related items