Performance Considerations

DiskSuite is not intended primarily to improve disk performance. It may do so in some cases, but it is also possible that a badly designed configuration will degrade performance. This section offers tips for getting good performance from DiskSuite.

Some performance considerations include the following:

• When concatenating, don't concatenate partitions on a single drive. For example, don't concatenate /dev/dsk/c0t1d0s4 and /dev/dsk/c0t1d0s7; you can do so, but it will hurt performance. If you want to concatenate these devices, make sure they are far apart in the list of devices in the md.tab file. For example:

d8 4 1 /dev/dsk/c0t1d0s4 1 /dev/dsk/c0t2d0s0 \ 1 /dev/dsk/c0t1d0s7 1 /dev/dsk/c0t0d0s0

will tend to have better performance than:

d8 4 1 /dev/dsk/c0t1d0s4 1 /dev/dsk/c0t1d0s7 \ 1 /dev/dsk/c0t2d0s0 1 /dev/dsk/c0t0d0s0

due to behavior of concatenations.

• When mirroring, don't have multiple copies of the data on the same drive. Always try to use at least two different drives for mirroring, since writes to the same drive contend for the same resources, and the failure of the one drive would mean the loss of all data.
• When logging UFS file systems, try to have the logging and master devices on different drives.

  However, if one of the disks or controllers in a metamirror is heavily loaded by other activity (for example, it contains another file system on another of its partitions), you may want to use the -r option. This mode will force the metadisk driver to read only from one partition, thus avoiding further overloading the other partition. (Note that the metadisk driver does not automatically choose the least loaded partition. You must specify this by making the less heavily-loaded metadevice the first submirror.)

• Try to have all drives in a metadevice on separate data paths. For SCSI drives, this means separate host adaptors. For IPI drives, this means separate controllers. By having the metadevice spread the I/O load over several controllers, performance and availability will be improved.
• Experiment with the read options of the metadisk driver in a metamirror. The default mode is to alternate reads in a round-robin fashion among the disks. This is the default because it tends to work best for UFS multi-user, multi-process activity.

• In some circumstances, using the -g option will improve performance by minimizing head motion and access times. This option is most effective when there is only one partition per disk, when only one process at a time is using the partition/file system, and when I/O patterns are highly sequential or when all accesses are reads.
• Don't mix disks or controllers of widely varying performance or technologies in a single metadevice or metamirror. Particularly in old SCSI or SMD storage devices, different models or brands of disk or controller can have widely varying performance. Mixing the different performance levels in a single metadevice or metamirror can cause performance to degrade significantly.

Availability Considerations

The metadisk driver's mirroring capability provides insurance against data loss with single point failures in a UNIX system. This section provides a few tips to ensure that you'll actually realize that safety.

Some considerations to be followed to ensure availability of data are:

• The /etc/opt/SUNwmd/mddb.cf or /etc/opt/SUNWmd/md.cf files should never be edited or removed.
• Make sure that the /etc/opt/SUNWmd/md.cf file is backed up frequently.
• When concatenating or striping components, mirror them if you have the spare disk capacity. Concatenating partitions reduces effective reliability somewhat, so mirroring any critical data stored in a concatenated metadevice is a good idea.
• When logging UFS file systems, mirror the logging device.
• When mirroring, keep the components of different submirrors on separate disks. Data protection is diminished considerably if components of two or more submirrors of the same mirrored metadevice are on the same disk.
• When mirroring, define the metadevices with components on separate controllers, if possible. Controllers and associated cables may fail more often than disks, so organize the submirrors of your metamirrors so that the submirrors are as independent as possible. This also helps performance.

• When creating RAID devices, define with components on separate controllers, if possible. Controllers and associated cables may fail more often than disks, so organize the submirrors of your metamirrors so that the submirrors are as independent as possible. This also helps performance.
• If the metadisk driver takes a metadevice offline, consider unmounting other file systems mounted on the disk where the failure occurred. Since each disk partition is independent, up to eight file systems may be mounted on a single disk. If the metadisk driver has "trapped" a failure (as indicated by metastat), other partitions on the same disk will likely experience failures soon. File systems mounted directly on disk partitions do not have the protection of metadisk driver error handling, and leaving such file systems mounted can leave you vulnerable to crashing the system and losing data.
• Minimize the amount of time you run with submirrors disabled or offline. During resyncing and online backup intervals, you do not have the full protection of mirroring. Your system administrator should run metastat regularly on any servers using the metadisk driver to check for metadevices that are not functioning properly.
• Do not mount file systems on an underlying component of a metadevice. If a physical partition is used for a metadevice of any kind, you must not mount that partition as a file system. If possible, unmount any physical device you intend to use as a metadevice before you activate it.
• After a component is defined as a metadevice and activated, do not use it for another purpose.
• Use the block device name (for example, /dev/md/dsk/d8) for block device functions such as mount(1M) and umount(1M); use the raw device name (/dev/md/rdsk/d8) for raw device functions such as newfs(1M), dd(1M), or fsck(1M).
• Stripes and concatenations cannot be defined using a metadevice as a component; only physical partitions can be used as components of a stripe or concatenation.
• For mirrored metadevices, only the metamirror can be mounted. An error message is returned if you try to mount a submirror directly, unless the submirror is offline and mounted read-only. However, you will not get an error message if you try to mount a physical partition that is a component of a metadevice; this could destroy data and crash the system.

• Two DiskSuite startup files are installed in system startup directories to effect automatic rebooting and resyncing of metadevices: /etc/rcS.d/S35SUNWmd.init and /etc/rc2.d/S95SUNWmd.sync. See Chapter 2, "Installation and Setup," for detailed instructions on setting up DiskSuite.
• If you want all swap space to be mirrored, you should use swap -l to check for swap devices. Partitions specified as swap must be mirrored separately.

Capacity Considerations

DiskSuite increases system capacity by supporting the concatenation, striping, or RAIDing of components into metadevices. These metadevices can then be used as either file systems or raw devices.

Some considerations that deal with capacity are:

• Components of differing physical size or geometry can be used effectively with the metadisk driver; what is important in defining metamirrors is that the mirrored metadevices be the same size, or nearly so. If you define a metamirror with two different sized metadevices, the metamirror will be the size of the smaller of the two metadevices; the extra space on the larger metadevice is unused.

  If you define a metamirror with one metadevice, the metamirror will be the size of that metadevice, regardless of the size of metadevices attached later. You will not be allowed to attach a smaller metadevice.

• When metadevice state database replicas are placed on a component used by a metadevice, the capacity of that metadevice is reduced. The space occupied by a replica is rounded up to the next cylinder boundary and this space is skipped by the metadevice.
• On components with labels (those starting at cylinder 0), the first cylinder of a component is skipped when the component is not the first component of a metadevice.
• All components of a stripe are the size of the smallest component. The components are rounded down to the nearest multiple of interlace size. So, if different size components are used within in a stripe, then disk capacity will be limited to a multiple of the smallest.
• When you're using the file system logging facility, remember that larger logs result in greater concurrency.

Labeled Partitions

All physical devices must have one disk label. The label is normally created by the install software, the format program, or the fmthard program. The label can appear on more than one of the logical partitions that are defined in the label.

Physical partitions that contain a label should not allow a user to write to the block that contains the label. Normally this is block 0. UNIX device drivers allow a user to overwrite this label.

Security Considerations

DiskSuite does not provide an audit trail for any reconfiguration of metadevices that may be performed on the system. This means that DiskSuite does not support C2 security.

Compatibility Considerations

DiskSuite is compatible with most other Sun Microsystem products. However, there are some limitations and considerations.

Some considerations that deal with compatibility are:

• Machines using Solstice DiskSuite 4.0 must be running the Solaris 2.1 release or a later Solaris 2.x release.
• To use the DiskSuite UFS file logging or diskset features, you must be running the Solaris 2.4 or a later Solaris 2.x release.
• When using Sun Prestoserve^(TM) on a system running DiskSuite, you should not use Prestoserve features on metamirrors or their submirrors; or on metatrans devices if the logging or master devices are mirrored.
• DiskSuite is compatible with the Online: Backup 2.0 unbundled product.
• The DiskSuite diskset feature is not supported on x86 systems.