VISA存储网络系统元数据管理的研究

Metadata management of VISA storage network systems

为了提高虚拟接口存储结构（VISA）存储网络系统的性能，设计实现了基于块级的元数据管理子系统．该系统的元数据管理子系统实现了物理设备和逻辑设备的块地址空间之间的映射，支持动态地址映射、动态块重分配和热块冗余技术，可以对I／O请求的路径和数据传输的路径完全控制，从而有效地改善了VISA存储网络系统的I／O性能，使数据布局更加合理，提高了存储管理的灵活性，初步实现了负载均衡，并且使VISA存储网络系统具备了容错能力和节点的动态增加能力．测试结果表明：在加入了基于块级的元数据管理功能带来相应开销的情况下，VISA存储网络系统原型与传统存储网络系统iSCSI＋Linux MD相比，传输延迟减少了8．2％～21．0％，传输吞吐率提高了8．7％～11．9％．

To enhance the performance of VISA （virtual interface based storage architecture） storage network system, a new metadata management subsystem is proposed. It is running at block-level. The high efficiency and scalability come from the following mechanisms： a clear-hierarchy mapping from physical to logical blocks, dynamic address mapping, dynamic block reallocation and hot block redundancy. Furthermore, the proposed subsystem can perform a full control over the data request and transfer path. VISA storage network system is improved with a more reasonable data layout, the flexibility of storage management and preliminary load balancing. It is also brought the functionality of fault-tolerant and the ability to add storage resource dynamically. The experiment result proves that, even though the proposed subsystem may bring more overhead, VISA still outperforms traditional storage network system such as iSCSI with Linux MD in both transfer delay and throughput, the increase is 8.2% to 21.0% and 8.7% to 11.9% respectively.