云原生存储无损扩容算法研究  

Research on Lossless Expansion Algorithm for Cloud Native Storage

作　　者：蔡旭辉 董晓荔 丁蔚然 陈曦 郑卿 CAI Xuhui;DONG Xiaoli;DING Weiran;CHEN Xi;ZHENG Qing(China Mobile Communications Group Co.,Ltd.,Beijing 100032,China)

机构地区：[1]中国移动通信集团有限公司,北京100032

出　　处：《数字通信世界》2024年第5期12-14,35,共4页Digital Communication World

摘　　要：分布式存储是多云数据中心的重要产品。云原生应用迭代周期缩短,对存储的容量和调整速度也有了更高的要求。目前,云原生存储因其自身架构的高可扩展性、健壮性和高性能等优点已占领主流市场,但集群扩缩容导致存储池长期处于数据迁移状态,引发了性能耗损高、集群表现不稳定等影响业务体验的问题。针对这些问题,该文提出一种优化算法,增加逻辑标签机制动态映射存储路径到物理介质,使得增量数据在扩容的存储介质落盘,避免大批量的数据迁移,让云原生存储可以在扩缩容场景下降低性能损耗和集群不稳定的概率。Distributed storage is a common and important product in multi-cloud data centers.In the cloud native era,the application iteration cycle is shortened,and the storage capacity and adjustment speed are also higher requirements.Cloud-native storage occupies the mainstream market due to its high scalability,robustness,and high performance of its architecture.However,cluster expansion and shrinkage lead to long-term data migration in the storage pool,leading to high performance loss,unstable cluster performance and other problems affecting business experience.This paper proposes an optimization algorithm by adding a logical label mechanism to dynamically map storage paths to physical media.In this way,incremental data can be dumped onto the expanded storage medium,avoiding mass data migration,and reducing the performance loss and cluster instability probability of the cloud native storage in the expansion or reduction scenario.

关 键 词：CRUSH 选择器 云原生存储 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]