BW－RAID系统的分布式异步版本识别机制

摘要

To solve the BW-RAID system’s problem of cache data version identification during its data redundant mode conversion from mirror to RAID4, a distributed and asynchronous version identification mechanism is presented . This version identification mechanism generates a new version for a logical block in a mirrored volume when its data are updated , and during the redundant mode conversion , checks whether this logic block ’ s data cached in two mir-roring notes are consistent by comparing the versions between mirroring nodes , and if consistent , makes the data move to the data volume , otherwise stores them to disk caches temporarily to guarantee system redundant consisten-cy .It was proved by experiments that the mechanism can identify consistent data for all blocks in any state , inclu-ding normal , degraded , and recovery states .The sequential writing tests showed it improved the average bandwidth up to 25.43%with the storage overhead less than 1% compared to cache synchronization mode .The open-mail workloads replay tests showed , the storage loads were less than 40%, and the blocks updated finished redundant conversion as each workload replay ended .The proposed mechanism is essential to guaranteeing redundant consis-tency while improving storage space utilization in BW-RAID.%为解决BW－RAID系统中数据冗余模式从镜像向RAID4转换时的节点间缓存数据版本识别问题，提出了一种分布式异步版本识别机制。该版本识别机制在镜像卷中的一个逻辑块被更新时，为其生成一个新版本；在冗余模式转换时，通过比较镜像节点间的版本判定某一逻辑块在两个镜像节点缓存的数据是否一致，如果数据一致将其迁移到数据存储卷，否则暂存入镜像节点各自的磁盘缓存以保证系统的冗余一致性。实验表明，该机制在系统正常、降级、故障恢复状态下均能准确、有效识别一致数据；顺序写评测中存储负载小于1％，平均带宽提升最高达25．43％；Open－mail负载重播评测中存储负载低于40％，应用重播结束时写更新数据均完成冗余模式转换，实现了提高空间利用率的目标。