Analysis by John Burke
The Disk Drive Shuffle (Part 1)
A manager reported this scenario: Currently, all the drives attached to our system are 2Gb models. "We want to replace all of them with 4Gb models on a one-for-one basis. How can I bet do this?" Steve Dirickson provided the following thorough checklist:
1. Run :SYSGEN and update the ID= values of the disks to the new values
2. Run BULDACCT.PUB.SYS with the '@' parameter. If you're not familiar with BULDACCT, just run it interactively and use the '%HELP' command to have it tell you about itself
3. Turn off UDCs for MANAGER.SYS that might cause problems on a restart
4. Make an SLT
5. Make another SLT
6. Verify at least one of the SLTs with CHECKSLT.MPEXL.TELESUP
7. Make a full system backup; i.e. :STORE /;*T;DIRECTORY (or
@.@.@ if you prefer)
8. Make another full system backup
9. :VSTORE one or both of the backups. Note that you can combine the SLT
with the full
system backup if you use the STORE= option of the :SYSGEN 'TAPE' command
with an
indirect file containing a line like
"/;DIRECTORY[;SHOW;PROGRESS;whatever other
options you like".
10. Shut down the system
11. Swap the drives. Make sure that the SCSI IDs of the new drives are set to match those of the drives they are replacing.
12. Start the system
13. ISL>INSTALL from the SLT
14. :RESTORE BULDJOB1.PUB.SYS
15. :STREAM BULDJOB1 to regenerate your account structure and
users
16. :RESTORE / (or @.@.@);OLDDATE to get all
your files back
17. :STREAM BULDJOB2 to re-attach your UDCs
If you have the Installation, Update, and Add-on Manual from your 5.0 installation or the PowerPatch Installation Manual from one of your PowerPatch packages, it will step you through the process, albeit in a much more verbose format. Use the "Reinstall" path, inserting the hardware swap between the save-everything and the put-it-back phases.
Gilles Schipper added "one slight nitpick":
I would replace steps 14 thru 17 with:
:hello manager.sys;hipri Your account structure and UDC environment will be restored
upon completion.
BULDACCT is not necessary -- unless your directory on tape was somehow
damaged --
but of course you have that available as a backup procedure as necessary.
:startspool lp
:file t;dev=tape
:restore *t;;directory;show
:restore *t;/;olddate;keep;show=offline;create
I want to emphasize the ";olddate;keep;show=offline;create" options of the above restore. The ";olddate" will prevent all file modification dates from being changed to today, which will cause your next partial backup to become a full backup and will essentially upset any future file management activities based on date-last-accessed.
The ";keep" option will prevent LEDV1-located files from being placed elsewhere -- besides, there is no need to re-restore files already in place. The ";show=offline" option will give you a hardcopy of the restore output - almost always useful - but most useful, I find, after you have omitted this option. The ";create" option may create new users on your system but will ensure that ALL files from your backup are restored. Of course, the extra users created should be then deleted - and the listing of these users is included in the listing resulting from the ";store=offline" option.
[Editor's note: Breaking up the restore above into two steps was questioned since theoretically it can be done as one step. Gilles responded that it was a combination of "old habits die hard" (there was a bug in 4.5 that would cause you great grief if you did it in one step) and intuition. I side with Gilles. Restoring the directory separately requires little extra time, yet gives you an extra measure of peace of mind. Actually, I go even further and create a tape with just the directory.]
Lars Appel added:
Just one item to add after step 13. You should check DSTAT ALL and use VOLUTIL SCRATCHVOL / NEWVOL to make the discs beyond LDEV 1 usable to the system (in your case it will only be LDEV 2). DSTAT should show discs in state MASTER or MEMBER before beginning the RESTORE. (VOLUTIL is needed unless you have defined VOLumes in SYSGEN IO config).
The Disk Drive Shuffle (Part 2)
This scenario and checklist were provided by Bill Lancaster, with additions by others as noted.
Scenario: I have a customer who was in serious need of some hardware and hardware setup re-design. They had 30 Gb of disk all in the system volume set and wanted to add 20 Gb more, going to multiple volume sets.
My checklist and some comments:
1. Perform two full system backups and verify each.
2. Create a new SYSGEN tape.
3. Check the new SYSGEN tape with CHECKSLT.
4. Copy @.@.SYS to a separate tape and verify.
5. Verify all disk drives configured and working properly.
6. Create a BULDACCT job for each new volume set with just the accounts destined for that volume set.
7. Verify that a current full BULDACCT exists on tape.
8. Shut down the system.
9. Restart the system.
10. From the ISL prompt, INSTALL.
11. In VOLUTIL, scratch all drives except for LDEV 1.
12. In VOLUTIL, do NEWVOL volset:member# LDEV# 100 100 for
each volume in the
system volume set (other than LDEV 1).
13. In VOLUTIL, do NEWSET volset member# LDEV# 100 100 for the
master volume for
each new set.
14. In VOLUTIL, do NEWVOL volset:member# LDEV# 100 100 for
each volume in each
new set.
15. Restore SYS account files with ;KEEP;SHOW;OLDDATE options.
16. Stream all BULDACCT jobs to create accounts structure.
17. Restore all files with ;KEEP;SHOW;OLDDATE options.
18. Spot-verify applications.
19. Once everything appears OK, run a BULDACCT.
20. Perform a full system backup.
Our project was a bit more involved in the we were also installing a DLT backup using Orbit software, as well as installing the new 4.3-Gb half-height drives into a Jamaica enclosure at a 5.0 PowerPatch 2 level.
Everything worked except Step 15 where we were going to restore from a separate tape with @.@.SYS. We had neglected to verify that tape and found that we had a bad tape. The reason we did this separate tape was for additional comfort-level, since we were relying on a third-party backup product. Call me paranoid but I like the feeling of knowing I can get at least the operating system back up without third-party intervention. As it turned out, I had to rely on the Orbit tapes anyway! The moral of this story is "Be sure to verify all tapes you will restore from!".
A few rules of thumb we followed:
1. Don't mix unlike-size drives on a single volume set. This is mostly an operational consideration. We wanted to avoid having the small volumes in the set fill up first with plenty of space left on the larger volumes.
2. Put the more critical user accounts on the faster, newer disk drives. In our case we had a mix of 1.3, 2.0, 4.0 and 4.3 disks so we had to pay attention to this.
3. Set up the volume sets in a business-logical manner. In other words, put accounts in a volume set with other, related accounts, if possible. In our case, this was simple since this customer is essentially a service bureau for many clients. We were able to clearly isolate the volume sets along client boundaries.
4. Since we were completely redesigning the disk environment and doing an INSTALL, we made certain to have at least two verified backups. With the exception of the problem already mentioned, this worked well.
5. Don't be afraid to have a volume set made up of drives configured on multiple controller paths. In our case, we had three single-ended I/O controller cards. On a few of the sets, we had drives from each.
Stan Sieler added:
On Step 6, remember that all the groups in a given account don't need to be on the same volume set. I have several accounts where the R&D oriented groups are on one volume set, and the production groups are on a different volume set. One example is our WWW account. I wanted the Web server to be available even if our USERS volume set wasn't up, so I put the important groups on MPEXL_SYSTEM_VOLUME_SET, and the utility files/source on a different volume set.
Step 11.1: For each drive, ask: Is this drive a 9 Gb (or bigger) drive? Am I on a release BEFORE MPE/iX 5.5 PowerPatch 2? If the answer to both questions is "yes", then stop immediately and update to 5.5 PowerPatch 2 before adding a 9Gb drive!
Step 14.1: Do a VSCLOSE on the volume set.
Step 14.2" Do a VSOPEN on the volume set (the really paranoid can replace steps 14.1 and 14.2 with a reboot!)
Step 14.3: Build some group (PVTEST.SYS, perhaps) on the volume set (any new group will do)
Step 14.4: For each member, #, of the volume set, do:
:build member#.pvtest.sys;dev=ldev#(where ldev# is the ldev that the volume is on)
:purge member#.pvtest.sys Step 14.5: purge the test group (PVTEST.SYS in this example)
Steps 14.1 through 14.5 double check that the volumes are completely accessible, and that the volume set information table (VSIT) on the master volume of the volume set was correctly posted to disk, and that sectors 0 and 1 for each volume were correctly posted to disk. I've seen a couple of cases where one or the other failed to happen, and the user didn't encounter a problem until the next time they rebooted. Of course, that shouldn't occur, but...
Finally, on note one, several people observed that both De-Frag/X from Lund Performance Solutions and Disk Space Manager (DSM) from Bradmark can evenly distribute files across all disks in a volume set. [Editor's note: see the July 1997 issue of the NewsWire for a descriptio n/comparison of De-Frag/X and DSM.]