Dynamic Resource Allocation And Fault Diagnosis Probing Method Considering Cost In Cloud Computing

The size and complexity of cloud computing is largely increasing with its popularity, which also brings challenge for management.With increasing size of cloud computing data centers, energy consumption becomes the main part of operation cost. Cloud computing has to hold large amount of computing resource to meet peak request. Most servers in data center are underutilized in most time but they also consume much energy. Dynamically allocation resources based on workload of data centers, to improve resource utilization and reducing energy consumption, is meaningful for cloud computing.Cloud computing depends on the support of computing network. While the large size and complexity of network brings larger cost of network management. Traditional passive event correlation is not practical in fault diagnosis of cloud computing network. Active probing method with its flexibility may be a good solution for fault diagnosis problem in cloud computing network. However cost of probing is limited as probing will also bring negative impact on network. Efficient probe selection method for active diagnosis is another point of this paper. The main contributions of this paper include:(1) To solve the high energy cost and low resource utilization problem of cloud computing data center, a dynamic energy saving resource allocation method named DRAMDT is proposed in this paper. Considering that big gap among deadlines of virtual machine instances on the same physical server could lead to a long time underutilization of the server, which wastes a lot of energy, DRAMDT algorithm group virtual machine instances with a time rotated virtual machine sub resource pool. Only virtual machine instances in the same group could be packed on the same physical server. Simulation result verified the efficiency and effectiveness.(2) Based on the property of information entropy, a lemma about information entropy gain of probes is proved in this paper. Then a simple approximate method for probe information entropy gain computing is proposed. An efficient probe selection algorithm for active diagnosis, named MEAP (More Efficient Active Probing), is proposed in this paper with this approximate method. Simulation experiment in this paper verified the efficiency and effectiveness MEAP algorithm.