| With the large-scale application of cloud computing services,cloud service reliability is widely concerned by cloud service providers and users,multiple fault-tolerant techniques are adopted to enhance the cloud service reliability.At present,cloud service reliability has two key problems to solve.The first problem originates from that the current virtual cluster allocation rarely considers the heterogeneous failure probablility of physical machines and switches,and risk assessment is not considered in the process of the virtual cluster allocation,which often leads to downtime.The second problem originates from that the current container consolidation schemes have not considered the reliability issue raised by power consumption and Service Level Agreement(SLA)violation rate.Therefore,the virtual machines and containers on the physical machine need to be redeployed to improve the resource utilization and cloud service reliability of cloud data center.Although these two problems have gained widely attention and obtained many research results,there are still some deficiencies in the risk-aware virtual cluster allocation,virtual allocation reallocation based on proactive fault tolerance,and predictable container consolidation.Therefore,this thesis has studied thoroughly these three problems of cloud service reliability by which following achievements are got.1)Since the current virtual cluster allocation approaches mainly focus on optimizing algorithm itself,the reliability problem associated with the high risk cost of virtual cluster allocation is rarely considered.Therefore,a risk-aware virtual cluster allocation approach based on biogeography theory is proposed to solve this problem.Firstly,the mapping of the virtual cluster allocation and biogeography-based optimization algorithm is realized according to their characteristics.Secondly,the risk cost is modeled using the built-up data-intensive application model.Finally,the optimal solution is obtained through the improved biogeography-based optimization algorithm,and improves the reliability of cloud data center network under the constraint of network resource consumption and execution time.2)Since the existing fault tolerance schemes rarely consider the proactive fault tolerance problem of coordination among multiple virtual machines which jointly complete a parallel application,that is,if these virtual machines are not coordinated,the final execution results may be wrong.To solve this problem,a cloud service reliability enhancement approach is proposed using proactive fault tolerance.Firstly,CPU temperature is modeled to anticipate a deteriorating physical machine.Then,the selection problem of the optimal target physical machines is modeled as an optimization problem which is solved by an improved particle swarm optimization algorithm.Finally,the efficiency and effectiveness of the proposed approach are evaluated by the experimental analysis and comparison.3)Although the power consumption and SLA violation rate of workload consolidation have got more attention,the current container consolidation schemes rarely consider the reliability issue raised by them.To solve this problem,this thesis researches the predictable container consolidation scheme of cloud data center.Firstly,the linear regression model is exploited to detect and predict working state of the physical machine based on the existing CPU utilization data.Then,according to the working state and distribution of virtual machines and containers,a container consolidation scheme is proposed to guide the adjustment of these virtual machines and containers.Finally,multiple physical machine selection approaches are exploited to evaluate the container consolidation scheme,and reduce the power consumption of cloud data center under complying with SLA. |
