Ceph is a for outstanding performance, reliability and scalability of unity designed, distributed file system
1. The number can be easily extended to PB capacity
2. Various workload performance (per input / output operations [IOPS] and bandwidth)
3. High reliability
However, these goals compete with each other (e.g., scalability, performance, or reduce or inhibit affect reliability). Ceph's design also includes the protection of a single point of failure fault tolerance, it is assumed that large-scale (PB-class storage) storage failure is a common phenomenon rather than the exception.
Its design does not assume a particular workload, but includes the ability to adapt to changing workloads, and provide the best performance. It uses POSIX compliance accomplish all these tasks, allowing it to rely on the current semantic POSIX application (Ceph targeted by improvement) of the transparent deployment.
Ceph is a high-performance, scalable, and no single point of distributed file storage system, is developed based on Sage A. Weil-based storage service offers the following three:
- Object Store (Object Storage), either by using Ceph library, using C, C ++, Java, Python , PHP code may access or store data on Restful gateway object form, compatible Swift Amazon's S3 and of OpenStack.
- A storage block (Block Storage), as a block device like a hard disk mounted directly.
- File System (File System), the same as Network File System mounts, compatible with the POSIX interface.
Ceph structure, provided by the object storage and LIBRADOS RADOSGW, block storage provided by the RBD, the file system provided by the CEPH FS, and RADOSGW, RBD, CEPH FS require LIBRADOS call interface, and ultimately stored in the form of objects RADOS in.
Ceph cluster node has three roles:
- Monitor, monitor the health of the cluster, the latest CRUSH map is sent to the client (that contains the current network topology)
- OSD, the object of maintenance on the node, responding to client requests, synchronized with other nodes OSD
- MDS, providing documents Metadata, if you do not use CephFS can not install
Ceph storage is distributed uniformly randomly dispersed in each node it after file division, using Ceph CRUSH algorithm to determine the storage location of the object, as long as the topology of the current cluster, Ceph client file can be directly calculated storage location, with direct access to file OSD-node communication without the need to ask the central node to get the file location, thus avoiding the risk of a single point.
More architecture of Ceph can be found in the official description: Architecture
Ceph is now a more mature storage systems, and is an ideal storage backend OpenStack, Hadoop can be used as storage backend, which involves comparing the Swift and the HDFS.
Swift and Ceph
Ceph and Swift written in Python written in C ++, the performance should be the Ceph dominant. But with Ceph different, Swift focused on the object store, as OpenStack components of one of the proven mass production practices, combined with OpenStack well, many people currently using Ceph block storage to provide OpenStack, but still use the Swift object storage provided.
Swift developers have written articles and contrast Ceph Swift: Ceph and Swift: Why Fighting WE are not.
Ceph and HDFS
Ceph contrast HDFS advantage of easy expansion, without a single point. HDFS is designed for such Hadoop cloud computing is born, inherent advantages in the off-line batch processing of large data, and Ceph is a generic real-time storage systems. Although Hadoop can be used as Ceph storage backend (according to the official tutorial Ceph life and death can not integrate himself wrote a simple steps: http://www.kai-zhang.com/cloud-computing/Running-Hadoop-on-CEPH / ), but perform computing tasks or performance slightly inferior to HDFS (about 30% slower time Haceph: Scalable Meta-a using Hadoop for the Data Management Ceph ).