基于带外存储虚拟化的逻辑卷高性能快照
Logical Volume High Performance Snapshot Based on Out-of-Band Storage Virtualization
-
摘要: 在逻辑卷级别实现快照既能屏蔽物理存储设备的异构性,又能为各种应用提供透明的数据保护支持.提出了一种基于带外存储虚拟化系统的高性能快照方法LVHPsnap,它通过可扩展的元数据组织策略,能够灵活使用多种类型的快照适应不同应用的需求;通过以Finesnap卷、Checkpoint卷两种类型快照卷构建混合快照链表和压缩快照索引数据实现了高性能快照,以高频度生成快照时系统能保持较高的性能和存储空间利用率;引入自适应快照机制,通过监控数据块修改频度,使自动生成快照的时间间隔随应用负载的变化动态调整,实现数据按需保护.基于真实工作负载和基准测试程序的实验表明,在构建混合快照链表的条件下,LVHPsnap的性能和存储空间利用率随Finesnap卷的比例增加而提高,例如当Finesnap卷与Checkpoint卷比例分别为4∶1和8∶1时,LVHPsnap的性能分别较集群虚拟化系统LVM2提高了109.45%和130.86%,存储空间使用量分别仅为LVM2系统的43.40%和31.35%.Abstract: Snapshot implemented at logical volume level adapts to different underlayer physic storage devices, and transparently provides data protection service for many kinds of upper layer applications. A scalable metadata managing strategy is proposed, and with this strategy, storage system can freely support more than one type of snapshot volume to meet different requirements of applications. By compacting snapshot index data and constructing composite snapshot volume list with Finesnap volume and Checkpoint volume, storage system maintains high performance and high resource utilization ratio while the time to recovery from snapshot volume being short. After adopting self-adaptive snapshot generation policy, modification to data block will be traced, and the time interval of generating a new snapshot volume can be dynamically shifted to follow the waving of application loads. The snapshot system LVHPsnap built on the above technologies performs good performance in representative experiments. For example, when the ratio of Finesnap volume to Checkpoint volume is 4 to 1 and 8 to 1, the performance of LVHPsnap rises by 109.45% and 130.45% as the performance of cluster virtualization system LVM2, and the used space of LVHPsnap is only 43.40% and 31.35% of LVM2.