# |
Item |
Comments |
DMF - Management |
1. |
Implement upward movement from tape to a DCM.
|
|
2. |
Visibility of and control over the DMF request queue to
reprioritise or cancel certain requests.
|
Recalls in particular
|
3. |
When merging a tape, clear the queued up copies if the HSPARSE
flag is cleared. Or some other method of canceling a tape merge.
|
Implemented at ISSP 3.0 via run_merge_stop.sh
|
4. |
Having actual filenames (and possibly paths) in the dmf database,
making bfid/filename linking easier.
|
Implement and release XFS Parent Inode Pointers
and have DMF take advantage of them
|
5. |
Implement space quotas for total storage, both online and offline.
Ideally both reporting, and enforcement of hard/soft limits.
Possibly integrate into xfs_quota.
|
|
6. |
The three dm*adm commands would be enhanced if they used the
readline package to provide command history and editing.
|
|
7. |
Avoid recently discovered bad spots on tape by flawing chunks
residing on those spots using the CAT database, automatically using
the second copy of that data where possible. Use the records of
those flaws to automatically repair those chunks later on.
|
|
8. |
Ability to dmmove chunks of data instead of whole files, for use
in recovery of potentially faulty tapes.
|
This is very similar to the previous item
|
9. |
The option to have load balancing of recalls across both VGs
containing copies of the migrated files.
|
Possibly implemented at ISSP 3.0
|
10. |
Ability to specify which VG to use for recall (and possibly
dmmove) requests.
|
Implemented at ISSP 3.0
|
11. |
Ability to specify a drive for dmatsnf/dmatread commands.
|
Implemented at ISSP 2.3
|
DMF - Performance |
12. |
Automatically prioritise DMF requests by type.
|
For example, in order:
- migrations when filesystem filling
- interactive explicit recalls
- other recalls
- migrations when filesystem not filling
|
13. |
Don't process explicit recalls (dmgets) in FIFO order, but rotate
around the requesters to prevent one user accidentally denying
service to others.
|
|
14. |
dmmove speed-up by implementing tape to tape I/O, analogous
to a socket-merge, and/or the ability to use multiple filesystems
for MOVE_FS.
|
|
15. |
Increase use of parallel threads inside DMF.
|
|
16. |
Improved handling of large numbers of hard links, especially
by xfsdump.
|
|
17. |
Improved handling of 100 million+ inode counts, both by xfsdump
and by DMF utilities which rely on inode scanning.
|
|
18. |
If a file is to be recalled from a tape which is currently being
used for migrations, the recall blocks until the VG has finished
writing to the tape which may take a long time. It would be
better if the write-child switched to a different tape.
|
Implemented at ISSP 3.0
|
DMF - Monitoring |
19. |
Performance statistics or trace points.
|
Possibly via PCP or dmfman
|
20. |
Option to force dmstat to refresh now, rather than in 2 minutes.
|
|
DMF - End-User |
21. |
Fix the bug in dmfind incorrectly finding dangling symbolic links
with the '-type f' option, and dmls issuing an error message when
encountering a dangling symlink.
|
|
22. |
Migration and recall of multiple files as a single group.
|
Expressed in many different ways:
- migration/recall of directories
- control over the ordering of recalls to increase proximity on tape
- "bucketing" small files into a collection, possibly using tar, which is then migrated
|
23. |
Add a parameter to dmget and dmput to recurse into a directory;
some users feel dmfind to be too clumsy.
|
|
24. |
dmget, dmput and the other user commands should be able to
accept the output from "find -print0", in a similar way to
"xargs -0".
|
Implemented at ISSP 3.0
|
25. |
Allow dmput of files by non-owners if they have read access;
this is necessary to prevent space quota problems for the owner.
|
|
26. |
A PAR file which is having a region recalled should have a state
of UNM. Currently, it stays as PAR which gives no indication
that activity is taking place. After the recall of the region
completes, the file should then be PAR or DUL as appropriate.
|
|
27. |
Extend the current Windows Explorer indication of offline/online
file status to cover the case where an NFS-mounted DMF filesystem
is re-exported via CIFS.
|
DMF server»NFS»Windows server» CIFS»Windows desktop
|
28. |
DMF client programs should support filesystems mounted via
SMB/CIFS.
|
|
OpenVault |
29. |
Force-release in OpenVault, like tmfrls in TMF.
|
|
30. |
Real-time tape position data from OpenVault, like tmstat in TMF.
|
|
31. |
Rotation of OpenVault log files so they are not removed at restart.
|
|
32. |
Decent man pages for OpenVault.
|
|