Thanks to the people who replied.
After reading the replies and some experimentations and considering the
requirements and spcifications of the harware/software, I finally decided
not to go this way, anyway, I can tell you that you can split a disk
(partition) AT THE HSZ level and use one piece from system A and the other
piece from system B at the same time, both systems being connected to the
HSZ40 via a shared SCSI bus. I tested for a whole day, doing lots of I/Os
to the partitions from bth systems at the same time and got no corruption
of data. BEWARE: Don't try to mount a WHOLE HSZ40 disk from more than one
system at the same time, you would trash your data !
I decided not to go this way because if you want to boot from the HSZ40,
you need to have a device configured on LUN 0, all my LUNS 0 are presently
used and it would be a pain to reorganize my storage, not to mention that
the partitionned disk would be very busy because 2 systems would be
accessing it.
Regarding the Advfs vs LSM discussion, someone pointed that you could use
LSM even if you use hardware raid, to increase availability and/or I/O
performance by using more than 1 SCSI bus and mirrorring accros devices
that are located on DIFFERENT SCSI BUS. In this case, if a single SCSI bus
fails (eg. SCSI adapter malfunction, controller down, etc..) you can still
access your data via the surviving SCSI bus. This can get quite complicated
to set-up and manage tough...
Here is the original post as well as some answers:
-----------------
Hello,
I am planing to move my systems disks to the HSZ40 controller. I am using a
shared SCSI bus and 2 HSZ40 in dual redundant config. Most of my important
storage is protected by RAID but my systems disks are not and it is a
weakness.
I don't have much disks left to play with, and I will have a hard time
going to my boss' office and tell him that I need ANOTHER storage shelf (I
can only put one more shelf in the rack and there is only one port left
available on the HSZ's). I have 3 4Gig drives left, and I need a couple of
these from time to time as a work area for big jobs.
I was considering putting a second internal disk in each of my alpha 2100
and copy the systems disk there from time to time (or mirror it with LSM),
but this would be a pain if ever one breaks, I will have to shut down the
system to get it fixed. And I removed LSM from the system because it was
way more complicated to use than harware raid, for a DU neophyte, I still
wonder what are the advantages of having ADVFS + LSM. I've got the ADVFS
utilities + hardware raid, and apart of software stripping + mirrorring, I
don't see what LSM offers.
Anyway, I am drifting from my questions...
To minimize disk usage, I was wondering if I could "split" a 4 gig drive in
two with HSOF 3.0 partitions, and copy the system disks (2 Gig each) to
these partitions. In case of a system disk failure, I would boot from an
HSZ40 partition.
I know that you cannot use the same HSZ40 device from two different systems
at the same time (you risk data corruption), but I wonder if by any chance,
it was possible to do so with partitions (I mean, machine X uses partition
A and machine Y uses partition B) ?
Also, is it possible to boot from an HSZ40 device ? I heard that you MUST
use a LUN 0 to boot, is this true ?
Provided I can use partitions to copy my systems disks (or I copy my system
disk to a whole disk), I was wondering what to do with SWAP partitions.
What happens when a DU system boots and can't find the disk unto which it's
primary (or any) swap partition is supposed to be ? Does it boot anyway and
ignore that swap space or does it hang there ?
Thanks.
-----------------------
>From alan:
re: Booting from an HSZ.
It depends on the console support in the particular system.
Many of the older systems only knew to look for LUN 0
devices as boot disks. This just meant that you couldn't
use a non-LUN zero device as a boot disk. If your console
can see all the device than you should be able to use any
as the boot disk.
re: Missing primary page/swap.
I think the system treats the inability to find the primary
page/swap space the same as it does when /sbin/swapdefault
is missing; it uses lazy page/swap space allocation. If
the system doesn't find any of the page/swap devices, then
thing will run normally until you run out of physical
memory. At that point processes will start dying as they
are killed off so others can page, which they can't and
the system memory will drop down to the steady usage of
physical memory. Not that it will likely be useful as
such a point...
re: HSZ partitions.
Never used them.
re: LSM.
In a configuration where you can split your I/O across
multiple SCSI busses, host based mirror and striping
can always perform better than controller based mirroring
and striping going over a single bus. Also, in such a
configuration you can spread mirroring over multiple
busses which prevents the bus to an array controller from
being a single point of failure.
------------------
After discovering that I was talking about partitionning the storage set at
the CONTROLLER LEVEL, not at the OS level:
Personally, I don't think using that feature is necessarily a good idea.
It looks like it's different to the host when it really isn't... you
have to watch the IO loads to both separately and as a whole. On
the otherhand, you will get IO rates for each unit so you can distinguish
the contribution of each by doing it that way. You also lose the ability
to 'share' the excess between filesets sharing a domain (e.g., in our
case the excess we keep in /var for dumps is available for /tmp and other
temporary consumers).
Regardless, I seriously doubt you could share HSZ partitions between
non-ASE hosts (it would technically work, but it wouldn't be supported
and would be more risk than I'd care to take with my data, particularly
the root partition). My view of the controller partitioning feature
is good if you _must_ have smaller "devices" which are raid backed,
but otherwise something that just adds needless complexity. There are,
of course, other valid opinions.
>I've already partitionned a 4 gig drive with the HSZ40 and created 2 UNITS
>that I mounted on the 2 machines connected to the HSZ40 (I mounted 1
>partition / machine) and did some stuff on the partitions. I _SEEMS_ to
>work just fine, but I wanted to confirm that I would have no problems with
>such a setup.
Are you running ASE?
Technically it is possible to have two hosts share independent units
behind a single HSZ (or redundant pair), we did that for 18 months...
but Digital doesn't support it. We had a series of unusual problems
which were Digital's fault (three hsz battery incidents, faulty kzpsa
a09 firmware, oracle database corruptions likely resolved by
OSF375-055). Several times we got the "unsupported configuration"
story from Digital in chasing these problems (we hit all of them
long before they acknowledged them, and -055 was developed from
corrupted disks we shipped back to engineering), so we finally
decided it wasn't worth it.
-kurt
------------------------------
You can't RAID 1 across SCSI buses. At this stage, if your SCSI bus fails,
your entire system is toast.
AdvFS is used when you want the machine to recover quickly from a major
crash. With UFS, if the machine crashes and restarts, your filesystems will
probably need an fsck, and so will probably not mount. With AdvFS,
journaling eliminates the need for an fsck.
In DECSafe environments, you use hardware striping and then mirror in LSM
across a dual SCSI bus, giving you SCSI bus redundancy.
Do you have two machines hitting the storage subsystem, or just the one?
>To minimize disk usage, I was wondering if I could "split" a 4 gig drive in
>two with HSOF 3.0 partitions, and copy the system disks (2 Gig each) to
>these partitions. In case of a system disk failure, I would boot from an
>HSZ40 partition.
Statistics say the when a system fails, it's first because of a disk, then
because of human error. Controllers fail very infrequently. By splitting a
4 GB drive, and copying your system disks there, your putting all your eggs
in one basket, which is the likeliest to break.
>I know that you cannot use the same HSZ40 device from two different systems
>at the same time (you risk data corruption), but I wonder if by any chance,
>it was possible to do so with partitions (I mean, machine X uses partition
>A and machine Y uses partition B) ?
Not sure about that one.
>Also, is it possible to boot from an HSZ40 device ? I heard that you MUST
>use a LUN 0 to boot, is this true ?
*ponder*
I think you can boot from an HSZ, as long as the disk is at LUN 0 on the
controller side.
>Provided I can use partitions to copy my systems disks (or I copy my system
>disk to a whole disk), I was wondering what to do with SWAP partitions.
>What happens when a DU system boots and can't find the disk unto which it's
>primary (or any) swap partition is supposed to be ? Does it boot anyway and
>ignore that swap space or does it hang there ?
No idea. I suspect it hangs or commits suicide. I'm pretty sure it needs
some swap on the boot disk, even if its 128MB. I'm doing something similar
to an 8200 in 10 days, and I'm keeping 1GB swap on another disk, and
leaving 128MB on the primary, so it's there at boot time.
I've recently (last 3 months) had to do similar juggling on 2100s with
HSZ's - if you can give me more config info - I'd be happy to suggest ideas.
rgds,
gunther
------------
Guy Dallaire
dallaire_at_total.net
"God only knows if god exists"
Received on Thu Apr 03 1997 - 17:48:34 NZST