面向新型非易失存储器的文件级磨损均衡机制

自旋转移力矩磁存储器(spin transfer torque random access memory, STTRAM)和磁阻式随机存储器(magnetic random access memory, MRAM)等新型存储器具有接近于DRAM的访问速度，是构建高性能外存系统和提高计算机系统性能的重要手段，但有限的写次数是其重要局限之一.设计了文件系统级磨损均衡机制，使用Hash函数分散文件在外存中的存储，避免在创建和删除文件时反复分配某些存储块,通过分配文件空间时选择写次数较低的存储块，避免写操作的集中；使用主动迁移策略，在外存系统I/O负载较低时主动迁移写次数较高的数据块，减少磨损均衡机制对I/O性能的影响.最后在开源的基于对象存储设备Open-osd上实现了面向新型存储器文件系统级磨损均衡机制的原型，使用存储系统通用测试工具filebench和postmark的多个通用数据集进行了测试与分析，验证了基于新型存储器的文件系统级磨损均衡机制能稳定地将存储块写次数差减少到原来的1/20左右，同时最高仅损失了6%的I/O性能和增加了0.5%的额外写操作，具有高效和稳定的特性.

File System Level Wear Leveling Mechanism for Non-Volatile Memory Based Storage

The access speed of STTRAM, MRAM and the other Non-volatile memory is close to that of DRAM. So they are very useful for high performance storage system and improving the performance of large computer system, but their limited write endurance is one of the most important limitations. We introduce file system level wear leveling technology for them. Using the Hash function to disperse files on the storage system, some blocks are avoided allocating repeatedly when creating and deleting files. The blocks with lower write count are chosen to avoid some blocks with centralized write operation when allocating space for file. An active heat data migration strategy is designed to reduce the I/O performance impaction of wear leveling mechanism. Finally, we implement the file system level wear leveling mechanism prototype based on the open source object-based storage device named Open-osd. Using Filebench, postmark and some trace are tested and analyzed, and the results show that the difference of write count between blocks is reduced to around one of the twentieth with original, and only reducing 6% system I/O performance and increasing about 05% amount of writing. It is verified that the file system level wear leveling mechanism is effective and stable.