Study Of Some Problems On Resource Allocation In Cloud Computing

Cloud computing provides resource integration and reallocation for heterogeneous platforms via virtualization technology and SOA(Service Oriented Architecture) framework. Since the resources are offered as the form of service, users and applications can access the resources by predefined service interfaces. Resources are integrated into a “resource pool” logically, hence, cloud computing system can allocate resources in more flexible way by adopting multiple allocation policies according to the demands of users and applications. In cloud systems, user tasks are executed in the virtual machines(VM), however, the number and resource demands of the users change over time, thus, the quantity of VMs and the resources used for VM creation are also time-variable. Therefore, how to exactly describe the change characteristics of the cloud, make an allocation for users in a fair manner to ensure the computational tasks can be finished, are the essential issues in cloud resource management.Moreover, it is significant to design an evaluation model for allocation fairness measurement, that can provide quantitative basis for the selection of resource allocation algorithms.To address these issues in cloud resource allocation, this paper includes the following aspects:(1) Three mathematical models are established to describe the change characteristics of cloud resources. Firstly, to address the issue of different virtual computing node(VCN) tasks having different resource demands, “Dynamic Resource Demand(DRD)” model is proposed. Secondly,Secondly, during the operation of cloud platform, the new tasks will continue to produce VCNs,and the finished tasks will destroy nodes, so the number of computing nodes in the platform can be changed during a period of time. Thus, it results in changes of the occupancy(or remaining)amount of the resources. To describe this feature, “Dynamic Node Number(DNN)” model is presented. Finally, with the increase in the amount of user tasks and the number of users, platform resources may be unable to meet the growing needs of the users, which will make users need to wait for free resources to perform their task. To solve this problem, service providers need to expand the system to meet the increasing resource demands of users. To describe this, ”Dynamic Cluster Size(DCS)” model is proposed.(2) A bottleneck resource oriented allocation algorithm, DBRF, is proposed, which prevents the system bottleneck resource from being exhausted in a short time. DBRF algorithm can predict the bottleneck resource, and control the amount of bottleneck to be allocated by introducing a scaling parameter λ. Thus, the iterations of the allocation algorithm can be increased, and the lack of bottleneck resource of some users caused due to excessive convergence of the algorithm can beavoided. Simulation results validate that the bottleneck resource can be divided into finer granularity, and the bottleneck depletion can be effectively slowed down. Moreover, DBRF ensures that more users can be assigned the bottleneck resources.(3) To address the issue of malicious occupancy of resources in cloud computing systems, a credit based enhanced fairness approach, Credit-based Dominant Resource Fairness(cb DRF), is proposed. In cb DRF, the credibility evaluation mechanism is introduced, which measures the resource utilization of the user. For malicious resources occupancy, the user will be punished according to its reputation value. cb DRF is proved to satisfy the four principles of fairness in literature [1], and it further proves that the method satisfies two new features, ”Release incentives” and ”Punitive allocation”(both of the features enhance the allocation fairness, and prevent users from malicious resource occupancy). Simulation results show that, cb DRF can effectively assess the use of resources, and restrain the malicious resource occupation.(4) To quantitatively evaluate the results of resource allocation, a fairness evaluation model for multiresource system, “Dynamic Fairness Evaluation(DFE)”, is proposed according to dynamic characteristics of cloud platform. In this model, the changes in the number of VCN and the resource demands are fully considered. Two parameters, time and probability, are introduced to establish a pair of sub-models: 1) dynamic resource demand model, and 2) dynamic node number model.With the two sub-models, DFE is finally proposed, which can evaluate the fairness of resource allocation in cloud system. In the experiments, three typical allocation algorithms(max-min,Avg and DRF), and a utility based algorithm(α-fairness) are adopted to validate the model. The results show that DFE is able to effectively reflect the fairness variation in dynamic resource environment, and can provide important reference and basis for the selection of allocation algorithm.