Research On The Key Technologies Of Fault Detection For Cloud Computing System

In recent years, the use of cloud computing is becoming more and more popular. The relatively centralized and professional methods of management have been applied in cloud computing system, which makes it work with high availability. However, a large number of malicious attacks and its own complexity and large scale still make a variety of hardware and software errors occur, and this may lead to parts or all of services unavailable. So fault detection for cloud computing system has become a hot research area.In this thesis, the key technologies which can increase the availability of cloud computing system is researched based on host security. What’s more, a three level system model—data processing-fault detection-fault recovery is designed so as to analyze data, monitor and recover the cloud computing system in real time.Firstly, the main intrusion methods and fault recovery technologies for cloud computing system are researched. Then on this basis, an improved FastICA algorithm is presented for data optimization processing in fault detection. The weighted correlation coefficient is adopted in the phase of albinism processing and the Newton’s iterative method is improved for third-order convergence, thus solving the problem of information loss and convergence rate.For the problem of fault detection, the deficiency of existing methods is analyzed, and a Hidden Markov Model based on simulated annealing algorithm is proposed. The simulated annealing algorithm is adopted to gain the initial parameters of hidden Markov Model, thus solving the problem of parameter sensitivity and global optimization. What’s more, the detection precision is improved.For the problem of fault recovery, the typical recovery techniques are analyzed and the coordinated checkpoint technology was studied which is suitable for this system. When a failure happened, a rollback recovery would be performed based on a globally uniform checkpoint.Lastly, the performance data is collected by sysstat in a virtual cloud computing system where the performance of every module in the system is verified. The experimental results show that this system can effectively reduce the data progressing time and information loss. What’s more, the accuracy is higher and false alarm rate is lower. So the system has good performance.