with data locations ● Requirements for elastic block storage ● Requirements for file systemopen-source storage ● Requirements ○ Cloud Native ○ Easy operation and maintenance ○ High performance ● Curve cannot be achieved without the support and participation of community contributors. Project address https://github.com/opencurve/curve Release cycle: a major release every six months, and a minor

to file /foo/bar in CurveBS 2. The address space of the block device /dev/sda maps to chunks of file in the system. For example, chunk1 corresponds to the address space of 0 to 4MB in /dev/sda, and and chunk2 corresponds to the address space of 4 to 8MB in /dev/sda. The size of chunk can be configured 3. Each file (/foo/bar) contains chunks scattered all over the storage nodes. ChunkServer structure is used to store user data, and user space data is transferred through reference of data address, avoiding user space data copy.

将conf中blockSize，chunkSize，metaServer和allocateServer ip保存在S3ClientAdaptor中 2.将conf中的S3相关信息：保罗ak，sk，s3address，bucketname等透传给S3Adapter模块。这里要注意，S3Adapter为原来curve块存储就有的模块，负责将快照数据上传到S3，这里我们对于s3的操作主要复用了这 个模块。该模块

like a duck, swims like a duck, and quacks like a duck, then it probably is a duck.业务系统 上游 Single Source of Truth 问题背景 数据孤岛怎么办？ 下游MySQL 怎么做 Binlog Maxwell DebeziumPolarDB-X 完全兼容 MySQL Binlog 可行性 • • Flink CDC 性能指标 • 25w rps • 8GB 大事务 • 5s 延迟* 下一步 • 验证更多工具 • GTID • 多流业务系统 上游 Single Source of Truth 问题背景 已有系统迁移怎么办？ 下游Demo for ReplicationPolarDB-X Replication 特性详情 提供与 MySQL 主备复制的能力 产品体验

Mds在分配空间时，轮流在不同的copyset中分配，每次从copyset中分配1个chunk， 这个chunk用copysetId:chunkId来唯一标识。COPYSET Copyset的生成策略：Source code : curve/src/mds/copyset/ bool GenCopyset(const ClusterInfo& cluster, int numCopysets, std::vector*