StorNext Storage Manager Known Issues

File retrieves from media to disk can be suboptimal for fast tape drives like the Oracle STK T10K drives. This scenario can occur when the retrieve event is initiated on a host that is different from the host running the mover process, which requires the use of synchronous direct I/O.

Workaround:

To work around this issue and achieve optimal performance for both file stores and retrieves with the T10K drives, increase the default I/O size used by the mover process and make the mover process use asynchronous buffered I/O when the use of synchronous direct I/O is not required, using the following steps:

Note: This workaround may also help improve the performance of the faster LTO drives like LTO-6, and LTO-7 by updating the FS_LTO_BLOCK_FACTOR sysparm.

Note: Changes to FS_xxx_BLOCK_FACTOR only affects tapes formatted after the change.

1. Change the FS_T10K_BLOCK_FACTOR sysparm from 8 to 32 by adding the following entry to
   
   /usr/adic/TSM/config/fs_sysparm_override:

FS_T10K_BLOCK_FACTOR=32;

Note: The T10K default I/O block size is 512 KB or 8 * 64 KB. With the block factor changed to 32, the new T10K I/O block size will be 2 MB or 32 * 64 KB. Presently, the FS_T10K_BLOCK_FACTOR sysparm must not be set to a value that exceeds 32.

2. Restart Storage Manager to ensure the change in Step 1 goes into effect:

# tsmstop

# tsmstart

3. Verify the FS_T10K_BLOCK_FACTOR sysparm contains the new value:

# showsysparm FS_T10K_BLOCK_FACTOR

FS_T10K_BLOCK_FACTOR=32

4. Save the current copies of your /etc/fstab on the MDCs and the DDM clients.
5. Modify /etc/fstab on the MDCs and the DDM clients to use the auto_dma_write_length and auto_dma_read_length mount options as follows:

snfs1 /stornext/snfs1 cvfs

rw,auto_dma_write_length=16m,auto_dma_read_length=16m 0 0

6. Unmount and re-mount your file systems.
7. Use new T10K media to store a copy of the file from the disk.

Note: Step 7 is very important; when the new copy is made to the new tapes, the new tapes are labeled with a 2 MB block size, which is used for subsequent writes or reads to and from the media. Tapes on which fsformat was run before the change will use the block factor in use at that time. This change will not impact those tapes.

Executing the command snbackup -s while a full or partial backup is running may result in a message that
/usr/adic/TSM/internal/locks/backup.lf is in an invalid format.

This is due to the snbackup -s process reading the backup.lf status file while the backup process is updating it.

Workaround:

Ignore the message; to clear-up the process, re-execute the command snbackup -s (provided that the backup is not writing to the backup.lf status file while snbackup -s is trying to read it again).

When copying files between media using the CLI command fsmedcopy, the file is not re-segmented to match the segment size of the destination media. Rather, the original segments are copied to the target media type and the distribution of segments across destination media will, therefore, be the same as the distribution on the source media.

Note: This behavior may cause file data segment distribution to be sub-optimal on the destination media.

Workaround:

Currently, a workaround does not exist for this Known Issue.

During StorNext Storage Manager start-up, SCSI Reservation conflicts may be reported by the IBM APFO driver if installed, particularly after involuntary fail-over of the StorNext MDC in High Availability configurations.

Workaround:

These reservation conflicts can generally be ignored and cause no issues with StorNext Storage Manager.