how to change clustersize of systemdisk

[Willem Grooters]

"Changing clustersize" for a datadisk is (I hope) staight forward: Create a new disk with different clustersize and BACKUP (without /IMAGE) all required data onto this disk. Right?
How do you do this with the system disk ?Problem is that all files, residing on [VMS$COMMON] are actually 'Entered' (SET FILE/ENTER) into the systemroot ([SYS0], [SYS1] etc) will be copies as files - not as directroy entries.
One thing I tried is to backup each directory separately, but still I ran into this problem with the system root(s).
Is there an easy way without having to redo all installation, and without having to install all layered products again?

[Dieter Rossbach]

There is no easy way (my opinion only) and yes, the links are the problem.

My first try og an answer:

1. Image-Backup of the system disk to a second one
2. set file/remove on disk2:[sys0]syscommon.dir
and on disk2:[sys1] ....
3. Init the new disk with the desired cluster size
4. Backup (without /image ...) disk2 to the new one
5. set file/enter=newdisk:[sys0]syscommon.dir newdisk:vms$common.dir
6. mc writeboot (perhaps)

Regards

Dieter

[Åge Rønning]

Why don't you use image backup?
Backup/image
Init with required clustersize + other init parameters
Backup/image/NOINIT

The restore command will not take init information from the backup, But use what you just used in the init command.

[Lokesh_2]

Hi ,

I also think like Age:

1. Take standalone backup of system disk [ or backup/image/ingore=(interlock,nobackup)]
2. Initialize your system disk with required cluster size
3. Backup/image/noinit ..

Thanks & regards,
Lokesh

[Dieter Rossbach]

I did a fast test:

$ init drb0: /struc=5/clu=26 newsys ! system disk is ods-5
$ mount /for drb0
$ backup/image/ignore=inter/noinit dra6: drb0:

and get:

%BACKUP-F-INVATTVAL, invalid attribute record value in save set

and yes: it works without /NOINIT .


Regards

Dieter

[Lokesh_2]

Hi Dieter,

Please remove /struc=5 from the following command - init drb0: /struc=5/clu=26 newsys.

That is use the following command, and let us know the results:
$init drb0:/clus=26 newsys

Thanks & regards,
Lokesh

[Hein van den Heuvel]

Why? What problem are you trying to solve?

Assuming I have extra disks I would like to keep my system disk clean and clear of any growing files through logical names. accountng, operator, sysuaf, rightslist,...

That way you basically do not have to make a backup of the system disk except for after major software changes.

fwiw,
Groetjes,
Hein.

[Antoniov.]

Hi
when I read Willen question I thought me because no IMAGE (I've used in tha past to backup & restore system disk); so I opened my VMS and typed help BACK/IMAGE and I read:

Directs BACKUP to process an entire volume or volume set.
Using this qualifier produces a functionally equivalent copy of the original volume or volume set.
[...]
Specifying the /IMAGE qualifier without also specifying /NOALIAS can result in incomplete disk or file restoration operations.
[...]
/------------------------------------------\
|Beginning in OpenVMS Version 7.2, when you |use the BACKUP/IMAGE command to restore or |copy a volume, the BACKUP utility preserves
|the cluster factor.
\------------------------------------------/
Before V7.2 was possible (so I known) change cluster factor as hint by AGe & Lockesh.
I think this limitation was introduced with new ODS-5 format to avoid some trouble.
So unique solution mey be the Ditier's answer.
Bye
Antoniov

[Dieter Rossbach]

@lokesh:

Backup/image won't work! I tried it, you get a "structure mismatch" error message! Remember: my system disk is ods-5,too.

Regards

Dieter

[Willem Grooters]

Hein:
Why? Simply, when I first started with this system I had only one (9Gb) disk, it was initilized to clustersize of 18 installing VMS. Since it was the only disk, I had to put all on that one. Now, new (small, as I'm a small user) disks were added and I want to clean up the system disk (indeed) and put it on a smaller clustersize ((much) less unusable space). Purely a matter of efficiency....It could also reside on the second disk, doesn't have to be restored on the original (I expect it even is the only way to get it copied ;-)) And it is an ODS-5 disk; a requirement, since Tomcat is on that disk as well.
Dieter: without /NOINIT I will be stuck to the big clustersize, I guess. Not what I want...
Lokesh (and others that suggested): why not making an image_copy/image/noinit to the second disk? Would do, would it?

[Hein van den Heuvel]

Just for grins....

Here is a perl script to parse DFU output to help guestimate the effect of cluster size on wasted space. Note: the script does NOT deal with 'overallocated' files, that is files which have more than a single cluster between the registred eof and the allocated space.

Sample output:

$ pipe dfu search DRA4: | perl cluster_size.p

1667 (1667) Files with 1667 headers 1983560/2014925 1.56%

Cluster Waste
------- --------- -----
1 0 0.00%
2 906 0.04%
3 1912 0.09%
4 3040 0.15%
5 4125 0.20%
:
100 134840 6.69%

# This perl script will process an DFU search output file
# to analyze the effect of cluster size on wasted allocaed space.
# use a PIPE or: defi/user sys$output dfu.tmp, and DFU SEARCH disk00:
#
# system$disk00:[directory]filename,ext;vers 1/9
#Primary headers : 6168
#Files found : 6168, Size : 2918870/3293460
while (<>) {
if (/(\d+)\/(\d+)$/) {
$end=$1;
$all=$2;
if (/Files found : (\d+)/) {
print "\n$1 ($files) Files with $headers headers $end/$all";
printf (" %5.2f%%\n", ($all-$end)*100/$all);
} else {
$files++;
$file{$end}++;
}
}
$headers = $1 if (/headers : (\d+)/);
}

#foreach $end (sort {$a <=> $b} keys %file) {
# printf ("%6d %d\n", $file{$end}, $end);
# }

print "\nCluster Waste\n------- --------- -----\n";
while ($c++ < 100) {
$waste = 0;
foreach $end (keys %file) {
$w = $c * int(($end +$c -1) / $c) - $end;
$waste += $file{$end}*$w;
# print "$c, $file{$end}, $end, $w, $waste\n";
}
printf ("%7d %9d %5.2f%%\n", $c, $waste, $waste*100/$all);
}



Enjoy...

Hein.

[Hein van den Heuvel]

Minor tweak on prior reply, now also reporting the number of files that would be mapped by a single cluster.

The script is in an attachment to avoid the forum sosftware from hosing the spaces.

Hein.

$ perl CLUSTER_SIZE.P < dfu.tmp

6168 (6168) Files with 6168 headers 2918870/3293460 11.37%

Cluster ____Waste _____ Single_Cluster
------- --------- ----- -------- -----
1 0 0.00% 1430 23.18%
2 3470 0.11% 1892 30.67%
3 7414 0.23% 2257 36.59%
4 10814 0.33% 2777 45.02%
5 14975 0.45% 3053 49.50%
6 18664 0.57% 3290 53.34%
7 22607 0.69% 3485 56.50%
8 27258 0.83% 3614 58.59%
9 31357 0.95% 3726 60.41%
:
99 508510 15.44% 5511 89.35%
100 513330 15.59% 5514 89.40%

[Åge Rønning · Accepted solution]

I just did the following on my system and it all seems to work ok. My current system disk is ODS2, But i converted the backup to ods5 and backed that up with /NOINIT without problems too. I'm running V7.3-1 + update patch.

init/clust=3 dkb300: tst
mou/for dkb300:
backup/image/noinit/ignore=interl sys$sysdevice: dkb300:

That gave me an ODS2 disk with different clustersize. Now, convert the old disk to ODS5 and make sure to INIT the new disk with STRUCT=5!

set vol/struct=5 dkb300:
init/STRUCT=5/clus=9 dkb200:
mou/for dkb200:
backup/image/noinit dkb300: dkb200:

That gave me a new ODS5 system disk with different clustersize.

Some time ago I saw a note saying you had to do a SET VOL/VOL_CHAR=HARDLINKS after the init of the new disk and before the backup/image/noinit, But I didn't do that.

[Willem Grooters]

Age,
What if the original disk is ODS5? I cannot backup to an ODS2-disk since I woul need to reinstall java and Tomcat (both require ODS5).
For all: Just wait a while posting solutions, please! There's quite enough to look at for the moment, but too little time (as usual too much to do and nothing to spare ;(. I'll grant the points later on.

[Willem Grooters]

I tried Age's suggestion:

$ init dkb300/cluster=3 A
$ mount/foreign dkb300
$ backup/noinit/ignore=interlock dka0: dkb300:

This is going right, on the running system I get quite a lot of warnings that flaes are opeend by another user - but it concerns logfiles. I don't care for that. At end I got the message %BACKUP-I-BTCROUT, Routine ODS-5 RMS SYNTAX ERROR.

$ DISMOUNT DKB300
$ MOUNT/OVERRIDE=ID DKB300

and now I see it's clustersize is, indeed, 3.

Files however have been copied with the original allocated size, into a multiple of 3. That's to be expected, due to /IMAGE, I guess. So next stap: how to get this allocated space free? For now I've a lot of files that are actually 2 blocks in size but 18 alloacted....
Oh yes - points have been granted so far.

[Hein van den Heuvel]

For all the plain SEQUENTIAL files you can simple use SET FILE/TRUNCATE.

$ copy/all=100 tt: tmp.tmp
aap
noot
mies
$ dir/size=all tmp.tmp
TMP.TMP;3 1/108
$ set file/trun tmp.tmp
$ dir/size=all tmp.tmp
TMP.TMP;3 1/9
$ type tmp.tmp
aap
noot
mies
$

[Hein van den Heuvel]

Arghh... that went out to quick.

I meant to add that this will ofcourse completely and utterly fragment you freespace. For a series of small files it will look like 3-used, 15 free, 3-used, 15 free,...
That would appear to be a good reason to defrag!

Hein.

[Lokesh_2]

Hi,

You can use DFU too. See the below thread

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=262626

Thanks & regards,
Lokesh

[Willem Grooters]

Hein,

>For all the plain SEQUENTIAL files you can simple use SET FILE/TRUNCATE.

Why this distinction? I mean: strictly sequential I can image clearly, but I don't understand why it wouldn't work with indexed and relative files. These have more allocated than used as well.
What about executables and libraries?

[Willem Grooters]

FYI: I was able to boot from DKA300; not that the system came up as it should, because a simple (stupid ;-)) reason: in some startup files there exists a hard-coded reference to DKA0: in stead of SYS$SYSDEVICE:. The disklabel wasn't changed by BACKUP either, so another thing to take in account...

[Hein van den Heuvel]

hein>> For all the plain SEQUENTIAL files you can simple use SET FILE/TRUNCATE.

willem> Why this distinction? I mean: strictly sequential I can image clearly, but I don't understand why it wouldn't work with indexed and relative files. These have more allocated than used as well.

Sorry... a bit late in replying. Well, basically the EOF for indexed files is not reliable and ignored by the RMS indexed file access code. It's just a hint, more or less maintained by convert and RMS. The truncate command knows this:
-SET-E-NOTRUNC, truncation not allowed for this file organization

I thought I remembered the EOF for relative file also to be sub-ordinate to a maximum record number, but SET FILE /TRUN does not complain for those. Still, I'd avoid mucking about with them. THey are pretty rarely used anyway no?


willem> What about executables and libraries?

No problem truncate as need be.


Hein.

[Jan van den Ende]

The trouble with BACKUP/NOINIT creating over-allocated files, and thereafter with afragmented disk after SET FILE/END, can be avoided very simple, easier, and quicker:
in the BACKUP command add; /TRUNCATE
It just looks like a plain, simple, straightforward, old-fashioned VMS-solution; and that is exactly what it is
;<)

[Willem Grooters]

Ok, might give a nice, unfragmented disk (if it works for indexed files as well). Keep that in mind.

BTW: How do I get rid of the scattered free space? DFU doesn't do defragmentation of free space. Just a matter of "COPY " and purge?

(bedankt Jan, had ik eerder moeten weten)