Journaling file systems can prevent data inconsistency problems caused by system crashes. Changes to the file system metadata (metadata) are saved in a separate log. When the system crashes, data can be restored correctly based on the log. In addition, the log does not need to check the file system when the system is restarted, thus shortening the recovery time.
1. Metadata is the data of the data. Data in any file system is divided into data and metadata. Data refers to the actual data in a common file, and Metadata refers to the system data used to describe the features of a file, such as access permissions, file owners, and distribution of file data blocks (inode ...) and so on. In the cluster file system, the distribution information includes the location of the file on the disk and the location of the disk in the cluster. You must obtain the metadata of a file before you can locate the file and obtain the file content or related attributes.
2. Metadata Management
There are two methods to manage metadata. Centralized management and distributed management. Centralized management refers to the management of metadata of a node in the system. All metadata is stored on the storage device of the node. Before all client requests to a file, you must request metadata for the Metadata Manager. Distributed management refers to the dynamic migration of metadata stored on any node of the system. Responsibilities for metadata management are also distributed to different nodes. Most cluster file systems use centralized metadata management. Centralized management is easy to implement, and consistent maintenance is easy. Therefore, satisfactory performance can be provided within a certain period of operation frequency. The disadvantage is the single failure point problem. If the server fails, the entire system will not work normally. In addition, when metadata operations are too frequent, centralized metadata management becomes the performance bottleneck of the entire system.
The advantage of distributed metadata management is that it solves the single failure point of centralized management, and the performance does not encounter bottlenecks as the operation frequency occurs. The disadvantage is that the implementation is complex, Consistency Maintenance is complex, and performance is affected.
[Original article link] http://blog.sina.com.cn/s/blog_6200c1440100w4mn.html
