| Cloud Computing, a new computing service scheme, represents an important paradigm of the information infrastructure for the consolidation and sharing of large scale of computing resources supported by state-of-the-art data centers that usually employ the virtualization technology. With the rapid development of Cloud Computing, an increasing number of enterprises and research institutes have deployed their applications in clouds. The great success of Cloud Computing in commerce attracts military interests, for which many military researchers and decision makers find that it is of great importance to apply the Cloud Computing technology to the C2 information systems. However, the military application of Cloud Computing still faces a serious problem, i.e., resource failures. Thereby, fault-tolerance becomes a significant requirement, especially for real-time tasks that abounds in C2 information systems.Fault-tolerant scheduling is an effective approach to realize fault-tolerance, among which the Primary-Backup(PB) model receives a great deal of attention. Up to now, some researchers have studied how to achieve highly efficient fault-tolerance in the traditional distributed computing systems. Unfortunately, little work has taken the features of clouds into consideration, neither can be applied to Cloud Computing. To address this problem, our work focuses on the features virtualized clouds, and studies the task scheduling and resource optimization methods based on the PB model. The main contributions of this work are as follows.(1) A comprehensive overlapping technique based on the PB model is proposedThe existing PB models are designed for the traditional distributed computing systems, and hence are not able to be applied to virtualized clouds. Our work extends the conventional PB model to meet the fault-tolerant requirement in virtualized clouds. In addition, a comprehensive overlapping technique is proposed which accommodates all the variants of the PB model to further reduce the consumption of backups, and so to improve the schedulability of systems.(2) The fault-tolerant constraints of VM migration are investigatedThe VM migration technique, an important resource management method in cloud data centers, is one of the factors to which the advantages of clouds can be attributed to. However, when the system employs the PB model to achieve fault-tolerance, the VM migration is confined by the fault-tolerant constraints. To guarantee fault-tolerance when VMs migrate across hosts, the constraints of VM migration are analyzed for the first time in our work.(3) A fault-tolerance oriented elastic resource provisioning mechanism is designedCloud systems are able to dynamically adjust the resource scale according to the resource request of tasks by the elastic resource provisioning mechanisms that are widely studied in previous works which, nonetheless, do not consider the fault-tolerant requirements. An updating method of VM cancel time in accordance with the PB model is proposed, based on which we design a fault-tolerance oriented elastic resource provisioning mechanism that scales up, or scales down, resources according to the task request and the status of resources to make clouds highly elastic.(4) An improved fault-tolerant scheduling algorithm for real-time tasks is proposedIn virtualized clouds, the basic computing instances are VMs rather than hosts, while the status of hosts also affect the performance of task scheduling. For this phenomenon, two scheduling strategies, I-AEAP and I-ALAP, are proposed by improving the classic fault-tolerant scheduling strategies, i.e., As Early As Possible and As Late As Possible. Based on the I-AEAP and I-ALAP, we design a novel fault-tolerant scheduling algorithm, named as FESTAL, using the above comprehensive overlapping technique, the VM migration technique, and the elastic resource provisioning mechanism. The experimental results show that FESTAL is able to effectively enhance the schedulability and resource utilization of virtualized clouds. |
PDF Full Text Request