Problem
I need more information on using the Virtual Disk Manager Release 1.0.0 on an SCO OpenServer 5.0.0 operating system.
Cause
RAID (Redundant Array of Inexpensive Disks) and the Virtual Disk Manager (VDM) are new features in SCO's OpenServer 5.0.0 product family.
This article is meant as a general introduction only and it assumes the reader is familiar with RAID terminology and theory. For further information, see the Operating System Administrator's Guide (hardcopy or online), the Performance Guide, and other articles in this database.
For brevity, SCO Virtual Disk Manager is referred to as "VDM" throughout this article.
This article is divided into 6 sections:
(1) Installation of VDM Release Supplement (2) Emergency Boot Floppy Set and Mirrored Root Disks (3) Creating Nested Virtual Disks (4) Adding New Pieces to a Disk Array (5) General Hints and Tips (6) References
Solution
Installation of VDM Release Supplement:The Release Supplement enclosed with the VDM product rectifies
some of the problems described in Chapter 2 of the SCO OpenServer
Features and Limitations manual. Specifically, the piece menu
and progress reporting problems are corrected.To install the Release Supplement, follow the instructions for loading and applying a software patch as described in Chapter 4 of the SCO OpenServer Handbook.
After installation of the VDM supplement on a remote source
server, if a subsequent network installation of the patched
VDM software is performed on a local client, a verification
error during installation is possible. If this occurs, you must
install and apply the Release Supplement directly on the local
client. A verification error will still occur, but it can
be safely ignored.Emergency boot floppy set and mirrored root disks.
If you have a mirrored root configuration, there are several important considerations when creating and using an emergency boot floppy set. If you create a mirrored root filesystem, create a new emergency boot floppy set immediately. If possible, create a system-specific boot disk rather than a generic one.
If you must create a generic boot disk (when the system-specific kernel does not fit on a floppy disk), then only the original piece of the mirror will open when the system is booted. You cannot use this boot disk on another system with a mirrored root.
When using the emergency boot floppy set to boot the system, use this bootstring instead of just pressing <Enter> at the boot: prompt:
fd(60)unix.Z root=vdisk(1) swap=none dump=noneIf root is not directed to vdisk(1) then the two halves of the mirror will become out-of-date, which could corrupt the filesystem when the machine is next booted from the hard disk.
Creating nested virtual disks.In general, virtual disks can be nested inside one another, provided that a virtual disk which has redundancy (mirrors and RAID 4/5) is not nested inside another virtual disk with redundancy, otherwise the virtual disk driver may hang, suspending access to all configured virtual disks. Three-way mirrors (created by nesting a mirror within a mirror) are an exception, and can be configured provided the master piece of the top level mirror is the second mirror. The following dktab(F) example shows the correct nesting needed to make a 3-way mirror:
/dev/dsk/vdisk3 mirror 2 32
/dev/dsk/1s1 2016 3200
/dev/dsk/2s1 2016 3200/dev/dsk/vdisk4 mirror 2 32
/dev/dsk/vdisk3 0 3200
/dev/dsk/3s1 2016 3200where "vdisk4" is the 3-way mirror. The nesting restriction will be lifted in a future Release Supplement.
When creating a nested virtual disk, VDM nests the disk definitions in /etc/dktab properly. If you edit /etc/dktab manually, make sure that definitions of virtual disks used as pieces occur before the definition of the virtual disk that references them.
Adding new pieces to a disk array.
When a new piece is added with VDM to an array with out-of-date
parity or a piece that is out-of-service, then a parity restore
will be be started automatically after the reconfiguration has
completed. This occurs in the background and is not shown by the
progress slider bar. The list of virtual disks is refreshed when
the progress slider bar reaches 100% and the reconfigured virtual
disk will be marked as having bad parity. When the restore has
completed and the display refresh interval has elapsed, the
reconfigured virtual disk will be shown with up-to-date parity
(or no pieces out-of-service).VDM does not allow adding disk pieces to an existing RAID 10 or
RAID 53 configuration. If you need to add another piece, create
the extra mirror or array piece with VDM, then exit and edit
/etc/dktab manually to add the extra piece. The format of this
file is described in dktab(F). You will need to move the
definition of the new virtual disk piece so that it occurs
before the definition of the virtual disk that uses it.General hints and tips for using the Virtual Disk Manager.
Maximum virtual disk size - You cannot create a virtual disk
larger than 50GB (gigabytes) using the VDM. You can create a
virtual disk up to 1 terabyte if the definition is created by
editing /etc/dktab with a text editor and then configured by
running dkconfig(ADM) on the command line.Create disks with an even number of blocks - If you create a virtual disk with an odd number of blocks, the last block (512 bytes) will be inaccessible through the block device node. When creating a filesystem on an odd-length virtual disk, mkfs(ADM) automatically rounds down the size accordingly.
Creating disks with parity data - In an array with parity
data (mirror, RAID 4, RAID 5) where whole disks are being
used as pieces in the array, the default piece size should
not be accepted. Some arrays give disk errors during parity
restoration and the virtual disk goes off-line. As a workaround,
select a piece size of about 5% smaller than the total disk
size.Piece limits - A virtual disk array can have up to 30 pieces defined, but this should ideally be kept below 20.
Cached partition and division nodes - VDM caches the partition
and division nodes (reads them at startup and holds them in
memory). If changes are made to these nodes while VDM is
running, you need to restart VDM so it can read the nodes again.Licensing - If licensing of VDM is not performed when the software is installed, then the License Manager will need to be invoked later to add the license data. After the License Manager has been run, the system will then need to be rebooted for the license to take affect.
Boot disk failure - When the root filesystem is mirrored, the parity disk cannot be used to boot the machine in the event of a boot disk failure. In this release, mirroring the root filesystem only protects against disk surface errors, not a total failure of the boot disk. A future supplement is scheduled to be released in the future which will address this and allow the parity disk to be used as a boot drive.
Mirror performance - There are performance problems with
mirrors in this release. In many cases the performance can
be increased by marking one piece of the mirror as the read
piece -- see the dktab(F) manual page for the correct syntax.
The performance is also scheduled to be addressed in a future
supplement.Checking the length of a piece - When a new virtual disk is
forced online with dkconfig -cf, the length of the pieces is
checked to see if they overrun the length of the device they
reside on. This checking is only performed when the -f flag
is used; other operations (such as reconfiguring a virtual
disk to change a piece or add a new piece) do not perform
this check.Parity restores - When a mirror or RAID 4 or 5 array is first
created, VDM states that parity must be restored before the
virtual disk can be used. This is incorrect. Restoring the
parity can be delayed until it is convenient and the virtual
disk can be used as normal. However, with out-of-date parity,
the virtual disk will go off-line if a disk error is encountered.Hardware requirements - Virtual disks can be composed of any
supported disk types and host adapters, and disks of any size
can be incorporated. In general, a RAID array will benefit
from fast disk access speeds and multiple host adapters with
large busses and on-board caching.Hot spares - A hot spare can be defined as another piece in an array, so if one disk fails the spare will be brought online and reconstructed from the data and parity on the working disks. Currently, there is a problem with the hot spare functionality. If the disk piece defined as the hot spare goes down for any reason, VDM will not be able to parse and initialize /etc/dktab properly, and will give a fatal error during the startup sequence. The disk array(s) will remain online, however.
Root and swap limitations - Are limited to RAID 1 (mirroring)
only; there is no striping available on root and swap. Other
space on the root disk not used for other filesystems is still
not configurable into disk arrays, since divvy information is
written to the partition. Other partitions on the root disk can
be used for virtual disk arrays, however. The boot filesystem
cannot be mirrored currently. This is scheduled to be addressed
in a future supplement. See also
Other articles in this database; SCO OpenServer 5.0 Operating System System Administration Guide; SCO OpenServer 5.0 Operating System Performance Guide; SCO OpenServer 5.0 Operating System Features and Limitations; SCO Virtual Disk Manager 1.0 Release Notes.