| Cloud storage is the fundamental component of cloud computing, its high reliability, high availability and high performance are the key guarantee to supporting all types of cloud services. Multi-datacenter cloud storage is built on WAN based distributed architecture, it ensure the reliability and availability of data service by multiple offsite copies in data centers around the world and ensure the performance of data by visiting the nearest nodes and downloading parallelly. Multi-datacenter cloud storage differs greatly from single-datacenter one in QoS guarantee and resource scheduling. How to guarantee different levels of service QoS and improve the system resource utilization is the key research area in cloud storage.In this dissertation, in order to guarantee the multi-datacenter QoS and improve the overall resource utilization of cloud storage, we studied the technique of optimizing multi-datacenter cloud storage resource scheduling to achieve QoS guarantee. According to the characteristics of multi-datacenter cloud storage, we divided the cloud storage service into four major sub-services which are load balancing, storage tiering, storage gateway and network transmission, and achieved the dual goals of QoS guarantees and system resource optimization for different sub-services by using specific means. The main achievements and innovations of this dissertation are in the following:(1) Analyzed the technical principles of multi-datacenter cloud storage resource management, concluded the research status and problems of QoS guarantees and resource scheduling of cloud storage in multi-datacenter architecture. We conducted a comprehensive analysis of cloud storage architecture and QoS guarantee mechanism and principles in load balancing, storage tiering, storage gateway and data center network, summarized the advantages and disadvantages of each method, pointed out the challenges of multi-datacenter cloud storage management and designed a resource optimization simulation platform. It is the theoretical and experimental basis for studing the multi-datacenter QoS guarantee and resource scheduling technique of cloud storage.(2) Proposed a QoS global optimal cloud storage multi-datacenter load balancing scheduling model and proposed a quotient space-based hierarchical load balancing scheduling algorithm (QBHLBSA). We analyzed the load balancing mechanism of cloud storage and its problems in multi-datacenter architecture, proposed a QoS global optimal cloud storage multi-datacenter load balancing scheduling model. The optimization objective is to guarantee the performance requirements of different QoS level applications and maximizing the resource utilization of each data center. According to the hierarchical management characteristics of multi-datacenter cloud storage, we proposed a quotient space-based hierarchical load balancing scheduling algorithm. The algorithm can schedule the storage load from coarse to fine granularity, with faster convergence time and avoid falling into local optimal values easily by using traditional algorithms. The simulation results showed that the algorithm can improve resource utilization and overall system throughput of cloud storage, guarantee the read and write performance requirements of high QoS level applications.(3) Proposed an object tiering storage system model for application layer QoS guarantee and proposed a pricing based automated cloud storage tiering algorithm (PBACST). In order to protect the performance requirements of high QoS level applications in cloud storage, we build an application layer QoS guarantee oriented object tiering storage system model. The optimization objective of model is maximizing the utilization of cloud storage high speed cache resource pool under the constraints of different applications’ QoS requirements, storage capacity and throughput. Based on the characteristics of multi-datacenter cloud storage, we proposed a pricing based automated tiering scheduling algorithm. The algorithm can make decision distributively, the components make the object tiering scheduling cooperatively. The simulation results showed that the algorithm can improve the utilization of high speed cache resource pool and guarantee the read and write performance requirements of high QoS level applications.(4) Proposed a QoS guaranteed cloud storage multi-datacenter task scheduling model and proposed a dynamic bandwidth allocation based real-time task scheduling algorithm (DBABRTSA). We analyzed the congestion problem when different QoS level application competing with limited bandwidth resources, proposed a QoS guaranteed multi-datacenter task scheduling model. The optimization objective of model is guaranteeing the performence requirements of high QoS level real-time application and improving the overall thoughtput of cloud storage. According to the hierarchical management characteristics of multi-datacenter cloud storage, we proposed a dynamic bandwidth allocation based real-time task scheduling algorithm. The algorithm can dynamically allocate task bandwidth according to application priorities. The simulation results showed that the algorithm can guarantee the read and write performance requirements of high QoS level applications and the proportional usage fairness of other QoS level applications, meanwhile improve the overall system thoughtput.(5) Proposed a QoS guaranteed inter-datacenter network traffic scheduling model and proposed a bi-level multi-swarm PSO based network traffic scheduling algorithm (BLMSPSOSA). We analyzed the uneven utilization problem of network link bandwidth in multi-datacenter cloud storage, proposed a QoS guaranteed inter-datacenter network traffic scheduling model. The optimization objective of model is guaranteeing the transmission performance of different QoS level data and maximizing the resource utilization of inter-datacenter network. According to the hierarchical management characteristics of multi-datacenter cloud storage, we proposed a bi-level multi-swarm PSO based network traffic scheduling algorithm. The algorithm has faster convergence time and can avoid falling into local optimal values easily by using traditional algorithms. The simulation results showed that the algorithm can improve the resource utilization of inter-datacenter network and guarantee the transmission requirements of high QoS level data.In conclusion, by optimizing the resource scheduling of different sub-services, we achieved the multi-datacenter QoS guarantee and effectively improved the overall resource utilization and throughput of cloud storage. |
PDF Full Text Request