Hi there,
I thought this was an easy one. Well, apparently the problem was (or
is) not as trivial as I thought.
Almost all responses contained a different opinion.
Just an example: Some say 'c' has to be the whole disk, some say it's not
mandatory but convention, so use it, and some say 'do anything you want'.
Anyway, I will try having 'a' start from cyl 0, use 'c' as the whole
disk and the rest of the partitions will be non-overlapping and of
the same size. If it doesn't work, I will let you know.
Just see for yourself. Here are all responses I received.
Thanks a lot to all of you:
pct_at_atom.ansto.gov.au
Gary Menna <G.Menna_at_isu.usyd.edu.au>
Phil Farrell <farrell_at_pangea.Stanford.EDU>
"Dr. Tom Blinn, 603-881-0646" <tpb_at_zk3.dec.com>
rwa_at_cs.athabascau.ca (Ross Alexander)
"Tim W. Janes" <janes_at_signal.dra.hmg.gb>
baranowskit_at_gdynia_at_genesis.prokom.gdynia.pl
IY_at_vilspa.esa.es
alan_at_nabeth.cxo.dec.com (Alan Rollow - Dr. File System's Home for
Wayward Inodes.) John Hascall <john_at_iastate.edu>
pct_at_atom.ansto.gov.au writes:
According to Juergen Bock:
.........
> Is that possible?
Yes, this is possible. (I haven't checked your figures.) We have two
drives where the only partitions which overlap is c with everything
else, ie excluding c there are no overlaps. We also set empty
partitions (size=0 offset=0) for those we don't use just to avoid
overlaps. I believe it is only conventions that make c the entire
drive and a start from offset 0, however I suspect there might be some
utilities that assume both partitions a and c start at offset 0.
Someone more knowledgeable will probably let you know.
By the way, you need only edit the size and offsets, disklabel will
fix the Cyls (which are actually a comment) and installing a file
system will fill in the others.
Note however that I have no experience with RAID, however I don't
think it is relevant here.
Gary Menna <G.Menna_at_isu.usyd.edu.au> writes:
Life was not meant to be difficult !
What I did was to create a whole bunch of LOGICAL disks
with the RCU. When It asks you to create a logical drive , change the
size of the capacity to what ever you want any hey presto you have
ANOTHER drive. Then just use the C partition. Much easier especially
when playing with databases.
Phil Farrell <farrell_at_pangea.Stanford.EDU> writes:
You can repartition as you please. Overlapping partitions cannot be
used simultaneously. I prefer to leave the C partition alone, because
some utilities like to use it to access the raw disk drive and it also
preserves the information about the entire disk. But that leaves me
the other seven partitions to play with. If I want the whole disk as
a single partition, I remake "a" with the same specs as "c" and use
"a". For two pieces, I divide the disk between "a" and "b", etc. I
usually set the unused partitions to size 0 and offset 0, so it
obvious that they are not being used.
"Dr. Tom Blinn, 603-881-0646" <tpb_at_zk3.dec.com> writes:
You can do ANYTHING at all that you want to do, as long as you never
try to use the disk as a system disk (which you shouldn't do anyway).
Bear in mind that if you ever lose the disk label information when
you've got a totally non-standard partitioning, you'll be totally
hosed, so it is a good idea to routinely save off the disk label
information to a file that is backed up.
Being that as it may..
If a disk is a system disk, then it is good practice for the "a"
partition to begin at block 0, since the system software knows about
the things that need to be at reserved places in the root partition on
a system disk (that is, a disk from which you will boot the system).
If you intend to use the "b" partition for swapping, it's a good idea
to not have it begin right at the beginning of the disk, although the
system software knows that the disk label will be in a reserved
location and will avoid over-writing it (which some people don't
understand -- at least, Digital UNIX is smarter than that although
some other UNIX implementations might not be). By convention, the "c"
partition is the entire drive, but it doesn't have to be. All of the
other partitions are entirely at your discretion anyway, although
Digital UNIX will, by convention, put the remaining partitions in
specific places and assign them specific sizes and cause certain
overlaps. If you want to do what you proposed in the balance of your
message for a disk that will be used as a "user" disk, it certainly
should work, with the warning that you need to be careful to save the
partition information where you can restore it if anything goes wrong.
rwa_at_cs.athabascau.ca (Ross Alexander)writes:
> does 'c' have to be the whole drive?
yes.
> Do 'g' and 'h' have to overlap 'd', 'e' and 'f'?
no.
> Or can I use all 8 partions available?
No. You can assign a, b, d, e, f, g, and h. C is *always* the whole
drive.
> Does 'a' necessarily have to use the first cylinder?
No.
> What I want is something like that:
> [lots of small non-overlapped partitions]
> Is that possible?
Yes.
"Tim W. Janes" <janes_at_signal.dra.hmg.gb> writes:
Here is our partition table on one of our RAID disks
8 partitions:
# size offset fstype [fsize bsize cpg]
a: 4000000 0 4.2BSD 1024 8192 16 # (Cyl. 0 -
976*) b: 2000000 4000000 4.2BSD 1024 8192 16 # (Cyl.
976*- 1464*) c: 16756736 0 unused 0 0 #
(Cyl. 0 - 4090) d: 2000000 6000000 4.2BSD 1024 8192
16 # (Cyl. 1464*- 1953*) e: 2000000 8000000 4.2BSD 1024
8192 16 # (Cyl. 1953*- 2441*) f: 2000000 10000000 4.2BSD
1024 8192 16 # (Cyl. 2441*- 2929*) g: 2000000 12000000
4.2BSD 1024 8192 16 # (Cyl. 2929*- 3417*) h: 2756736
14000000 4.2BSD 1024 8192 16 # (Cyl. 3417*- 4090*)
I just assumed that nothing needed to overlap and tried it. It has
been running for about 6 months without problems.
I think I tried to use 'c' but disklabel would not let me change its
size - but I am not sure.
baranowskit_at_gdynia_at_genesis.prokom.gdynia.pl writes:
I suggest you to install LSM (included in standard 3.2c set)
Then you should use raw "logical" volumes which can split acros
any number of physical devices. In example you can create one big
volume on all of your physical disks.
By that, LSM raw volumes can be greater than the biggest of your disk,
and they are faster (when splitted in multiple physical devices).
Generally: don't touch partition "C" - it must be all drive. There are
many
programs which assumes that.
All other partitions you can use how you like. Example
which you send in first look is correct.
IY_at_vilspa.esa.es writes:
You may partition the disk as in your proposal. c must be the
whole
drive (it is a convention).
You must be very careful not to mount two overlapping partition.
It is quite common to have overlapping partitions, in order to have
alternatives without re-partitioning the drive. In any case, a
re-format in obliged. But never two overlapping partitions should be
mounted (I think the operating system warns you against that).
alan_at_nabeth.cxo.dec.com (Alan Rollow - Dr. File System's Home for
Wayward Inodes.) writes:
The only requirement I know of is that you must have two partitions
that start at the beginning of the disk. I think disklabel assumes
that C does. And most people would be the least surprised if A did as
well. The reason you need two is that you can't change the size of an
open partition and to change one of the two you need other to remain
stable.
If you need more partitions than are easily available with
partition tables have you considered LSM? The two features
that come free with LSM and the ability to concatenate devices
and carve large devices up into smaller slices. I think you
can have upwards to 256 volumes out of a single device. LSM
may introduce more latency than a partition table and it
makes the system slightly harder to manage, but if you need
lots of little slices, it could be the better solution.
John Hascall <john_at_iastate.edu> writes:
The SWXCR driver/controller/something only allows *4* partitions, "a",
"b", "c", & "g", to be used. You can use all 4 (you need not follow
the convention that 'c' is the complete drive), for example:
root_at_afs-1# df -k
Filesystem 1024-blocks Used Avail Capacity Mounted on
/dev/rz0a 63567 42628 14582 75% / /dev/rz0g
1017327 395099 520495 43% /usr /proc
0 0 0 100% /proc /dev/re0a 6234605
3561995 2049149 63% /vicepa /dev/re1a 6234605
2806683 2804461 50% /vicepb /dev/re2a 6234605
3182415 2428729 57% /vicepc /dev/re0b 6234605
3668406 1942738 65% /vicepd /dev/re1b 6234605
4206688 1404456 75% /vicepe /dev/re2b 6234605
2721182 2889962 48% /vicepf /dev/re0c 6234605
3791653 1819491 68% /vicepg /dev/re1c 6234605
3954436 1656708 70% /viceph /dev/re2c 6188525
4824214 745458 87% /vicepi /dev/re0g 6234605
3811170 1799974 68% /vicepj /dev/re1g 6188525
5041761 527911 91% /vicepk /dev/re2g 6234605
3672760 1938384 65% /vicepl /dev/re8c 8128492
297 7315345 0% /vicepm /dev/re9c 8128492 79
7315563 0% /vicepn
This is my original post:
Hi everybody,
I assume this is kind of basic and has been discussed a bazillion
times. So please do NOT kill me.
We have to partiton two RAID drives with 7 x RZ28 and 4 x Quantum
Atlas. What I don't know (and can't find it either) is which partition
MUST overlap other partitions. I would like to have as many rew
partitions as possible (advfs won't do, I need raw partitions for a
database). So, does 'c' have to be the whole drive? Do 'g' and 'h'
have to overlap 'd', 'e' and 'f'? Or can I use all 8 partions
available? Does 'a' necessarily have to use the first cylinder?
The standard disklabel for 're'-drives gives me
# size offset fstype [fsize bsize cpg]
a: 131072 0 4.2BSD 1024 8192 16 # (Cyl. 0 -
31) b: 262144 131072 unused 1024 8192 # (Cyl. 32
- 95) c: 25196544 0 unused 1024 8192 # (Cyl.
0 - 6151*) d: 0 0 unused 1024 8192 #
(Cyl. 0 - -1) e: 0 0 unused 1024 8192
# (Cyl. 0 - -1) f: 0 0 unused 1024 8192
# (Cyl. 0 - -1) g: 24803328 393216 unused 1024 8192
# (Cyl. 96 - 6151*) h: 0 0 unused 1024
8192 # (Cyl. 0 - -1)
What I want is something like that:
# size offset fstype [fsize bsize cpg]
a: 3600384 0 unused 1024 8192 16 # (Cyl. 1 -
878) b: 3600384 3600384 unused 1024 8192 # (Cyl. 879 -
1756) c: 25196544 0 unused 1024 8192 # (Cyl.
0 - 6151*) d: 3600384 7200768 unused 1024 8192 # (Cyl.
1757 - 2634*) e: 3600384 10801152 unused 1024 8192 # (Cyl.
2635-3512) f: 3600384 14401536 unused 1024 8192 # (Cyl.
3513-4390) g: 3600384 18001920 unused 1024 8192 # (Cyl.
4391-5268) h: 3612672 21602304 unused 1024 8192 # (Cyl
5269-6151)
Is that possible?
**********************************************************************
Die Deutsche Bibliothek Juergen Bock
Zeppelinallee 4-8 069-7566-735
60325 Frankfurt bock_at_dbf.ddb.de
**********************************************************************
Received on Wed Sep 04 1996 - 11:22:37 NZST