Using IBM Spectrum Protect for Space Management (HSM)

IBM Spectrum Protect for Space Management (HSM) can be used on the AFM filesets or on the home.

The following figure illustrates HSM connected to home.

Figure 1. Sample setup of HSM connected to home
Sample setup of HSM connected to home
A new file created at home becomes candidate for migration to the IBM Spectrum Protect server. When a migrated file at home is read from cache, the file is recalled at home and served to the cache. If the cache does a write on a migrated file at home, the file is recalled at home and written to when the AFM queue is flushed. If multiple migrated files are written at the same time, the home issues recalls for all files at the same time. It is recommended that you exclude frequently changing files from HSM migration process to avoid recalls.

The following figure illustrates HSM connected to both home and cache.

Figure 2. HSM connected to both home and cache
HSM connected to both home and cache
When using HSM on an AFM fileset, the flag AFMSKIPUNCACHEDFILES must be set in the dsm.sys configuration file (IBM Spectrum Protect-related) to yes. For example - AFMSKIPUNCACHEDFILES yes. This parameter should be used for read-write cache filesets. It prevents the migration of dirty and uncached files. If this flag is not set, it might result in long waiters or unexpected cache states and unexpected errors in terms of the HSM migration processing. In the LU mode when this parameter is set and a file is made local because of updating data, migration of the file to tape might be prevented. For migration in the LU mode, do not unset this parameter. Migrated files cannot be evicted. File operations such as read, write, truncate, append or rename on migrated files recalls the files into the cache. When multiple migrated files are modified at the same time all recalls are submitted at the same time, IBM® recommends that you exclude frequently changing files or frequently read files from HSM migration process on both cache and home to avoid recalls.

It is recommended the following guidelines while using IBM Spectrum Scale AFM and IBM Spectrum Protect:

  • Prevent cache eviction in combination with IBM Spectrum Protect on the same fileset. Both techniques have the same goal to reduce the space required in the fileset. The combination of both techniques unnecessarily increases the complexity of the environment.
  • Start of changeIBM Spectrum Scale snapshots and IBM Spectrum Protect have a limited compatibility. The deletion of a stub file that is reflected in a snapshot (the snapshot was generated before or after the file was migrated) causes the recall of the file data. The file data is stored in the snapshot so that it can be accessed later. Therefore, do not use snapshots for an AFM fileset (in home or cache) and in the file system hosting the AFM fileset, if you are using HSM.End of change
  • When using IBM Spectrum Protect on home or cache be aware that access (read or write) to multiple migrated files at the same time causes bulk recalls. Access to multiple files can be caused by users such as when they copy an entire directory or by AFM when changed files are replicated to home where the previous versions are migrated. You can avoid these issues by using the optimized tape recall process, which requires a list of files to be recalled before processing.

When running IBM Spectrum Scale AFM and IBM Spectrum Protect backup operations, prevent cache eviction in combination with IBM Spectrum Protect backup on the same fileset, if possible. Evicted (uncached) files will be skipped from backup processing. This might lead to errors in terms of the versioning of files on the IBM Spectrum Protect server.

For detailed description about the setup and configuration of HSM for AFM, see Configuring IBM Spectrum Scale Active File Management.