How is Metadata of a File System Stored?

The OceanStor 9000 organizes metadata using dynamic subtrees. All metadata in a namespace are grouped based on name subtrees. Each name subtree is allocated to one MDS for processing. One MDS can manage multiple subtrees. The normal processing procedure is as follows: A CA sends a request to any MDS. The MDS queries the local dynamic subtree table. If the local name subtree is in charge of the request, the MDS processes the request. Otherwise, the MDS forwards the request to the MDS whose name subtree is in charge of the request. If such an MDS cannot be found, the MDS forwards the request to the MDS whose name subtree is most likely to be in charge of the request to find out the metadata. If one MDS fails, other MDSs will share the name subtrees managed by the failed MDS. If the number of MDS arrays increases or reduces or name subtrees managed by an MDS are frequently or seldom accessed, a load balancing policy is implemented to migrate some of the frequently accessed name subtrees to a seldom accessed node.

The latest version supports up to two levels of subtrees to lighten workloads. In this manner, metadata of a file is processed by two MDSs at most.

Hello zj5000!

Have a nice day!

The OceanStor 9000 stores metadata in the same way as data is stored. That is, metadata is distributed on multiple nodes. Metadata of the OceanStor 9000 distributed file system is stored by mirroring. The number of mirrors is automatically set to M+2 for the metadata of a file that is under N+M protection, making metadata more reliable than data.

The OceanStor 9000 stores multiple copies of each metadata strip. Metadata can be read and written as long as any one copy is intact. When data restoration is implemented, lost mirror data is copied from intact data.

Any further questions, let us know!