parameter * indicates by how much the lookup count should be decreased. * * Inodes with a non-zero lookup count may receive request from * the kernel even after calls to unlink, rmdir or (when requests properly and it is recommended to defer removal * of the inode until the lookup count reaches zero. Calls to * unlink, rmdir or rename will be followed closely by forget * unless the file or directory fuse_entry_param above. * * On unmount the lookup count for all inodes implicitly drops * to zero. It is not guaranteed that the file system will * receive corresponding forget messages for the

1 基于SPDK的CurveBS PFS存储引擎10/17/22 2 Why ●为了减少使用cpu做内存copy，减少系统调用 ●发挥某些被操作系统屏蔽的功能，例如nvme write zero ●根据阿里《When Cloud Storage Meets RDMA》的说法 ●在100Gbps网络带宽时，内存带宽成为瓶颈 ●Intel Memory Latency Checker (MLC)测试得到的CPU内存带宽是 ●应用程序不再通过系统调用在内核和用户态来回切换10/17/22 4 磁盘的读写 ●基于EXT4的存储引擎，依然需要通过系统调用来回切换 ●读写都需要CPU拷贝数据 ●不能发挥某些NVME的功能，例如write zero10/17/22 5 为什么用PFS ●对代码比较熟悉 ●找一个能管理裸盘，具有产品级可靠性的代码挺难的 ●PFS支持类POSIX文件的接口，与使用EXT4的存储引擎代码很像， 所以容易移植现有代码到PFS存储引擎 写PRP页面对齐内存分配代码10/17/22 11 pfs_pwrite_zero ●在初始化curvebs时，需要创建chunk pool, 每一个chunk都要填零 ●chunk不再被卷使用时，需要回归chunk pool，为了安全也需要填0。 ●使用nvme的时候，可以直接使用nvme write zero命令，不需要传递 大块数据（全是0），减少了nvme传输带宽，而且nvme在垃圾回收上

● Raft Consistency protocol High performance ● pre-created file pool ● data strip like RAID ● Zero data copy ● RDMA Cloud NativeCluster topology The physical pool is used to physically isolate belong to different failure domainsCURVE IO data flowOther performance optimizations RAFT protocol Zero data copy pre-created file poolCURVE file system File service middleware Upper-layer applications allocation algorithm with topology-based failure domain to provide high availability/reliability ● Zero copy; Data stripe; RDMA to improve performance ● Management and monitor tools ● Support CSI deriver

of the write buffer */ unsigned max_write; /** * Maximum size of read requests. A value of zero indicates no * limit. However, even if the filesystem does not specify a * limit, the maximum Asynchronous direct I/O requests * * Read-ahead requests are generated (if max_readahead is * non-zero) by the kernel to preemptively fill its caches * when it anticipates that userspace will soon read crc32c(uuid+id+xattrblock) */ /* id = inum if refcount=1, blknum otherwise */ __u32 h_reserved[3]; /* zero right now */ }; struct ext4_xattr_entry { __u8 e_name_len; /* length of name */ __u8 e_name_index;

pthread) for scalability and performance on Multi-thread CPU • Lock free queue design • Memory zero copy design • Cloud native supportCloud native for CurveBS • CSI plugin for CurveBS • Deploy

by 13% in the case of two replicas. The gap is even greater under high stress. 2. User space Zero copy of user data Iobuf structure is used to store user data, and user space data is transferred

Available in ->poll. Only set on kernels which support it. If unsupported, this field is set to zero. */ uint32_t poll_events; }; // fastcfs typedef struct fcfs_api_file_info { FCFSAPIContext