System Setup Guidelines

The hot-relocation feature is enabled by default only if you have the license for this feature. Although it is possible to disable hot-relocation, it is advisable to leave it enabled.

Although hot-relocation does not require you to designate disks as spares, you can designate at least one disk as a spare within each disk group. This gives you some control over which disks are used for relocation. If no spares exist, Volume Manager uses any available free space within the disk group. When free space is used for relocation purposes, it is possible to have performance degradation after the relocation.

After hot-relocation occurs, you can designate one or more additional disks as spares to augment the spare space (some of the original spare space may be occupied by relocated subdisks).

If a given disk group spans multiple controllers and has more than one spare disk, you can set up the spare disks on different controllers (in case one of the controllers fails).

For a mirrored volume, the disk group must have at least one disk that does not already contain a mirror of the volume. This disk should either be a spare disk with some available space or a regular disk with some free space and the disk is not excluded from hot-relocation use.

For a mirrored and striped volume, the disk group must have at least one disk that does not already contain one of the mirrors of the volume or another subdisk in the striped plex. This disk should either be a spare disk with some available space or a regular disk with some free space and the disk is not excluded from hot-relocation use.

For a RAID-5 volume, the disk group must have at least one disk that does not already contain the RAID-5 plex (or one of its log plexes) of the volume. This disk should either be a spare disk with some available space or a regular disk with some free space and the disk is not excluded from hot-relocation use.

If a mirrored volume has a DRL log subdisk as part of its data plex, that plex cannot be relocated. You can place log subdisks in plexes that contain no data (log plexes).

Hot-relocation does not guarantee that it preserves the original performance characteristics or data layout. You can examine the location(s) of the newly-relocated subdisk(s) and determine whether they should be relocated to more suitable disks to regain the original performance benefits.

Hot-relocation is capable of creating a new mirror of the root disk if the root disk is mirrored and it fails. The rootdg disk group should therefore contain sufficient contiguous spare or free space to accommodate the volumes on the root disk (rootvol and swapvol require contiguous disk space).

Although it is possible to build VxVM objects on spare disks (using vxmake or the Storage Administrator interface), it is preferable to use spare disks for hot-relocation only.

Do not place more than one column of a striped plex on the same physical disk.

Calculate stripe unit sizes carefully. In general, a moderate stripe unit size (for example, 64 kilobytes, which is also the default used by the vxassist command) is recommended. If it is not feasible to set the stripe unit size to the track size, and you do not know the application I/O pattern, it is recommended that you use 64 kilobytes for the stripe unit size.

	NOTE: Many modern disk drives have "variable geometry," which means that the track size differs between cylinders (i.e., outer disk tracks have more sectors than inner tracks). It is therefore not always appropriate to use the track size as the stripe unit size. For these drives, use a moderate stripe unit size (such as 64 kilobytes), unless you know the I/O pattern of the application.

Volumes with small stripe unit sizes can exhibit poor sequential I/O latency if the disks do not have synchronized spindles. Generally, striping over non-spindle-synched disks performs better if used with larger stripe unit sizes and multi-threaded, or largely asynchronous, random I/O streams.

Typically, the greater the number of physical disks in the stripe, the greater the improvement in I/O performance; however, this reduces the effective mean time between failures of the volume. If this is an issue, striping can be combined with mirroring to provide a high-performance volume with improved reliability.

If only one plex of a mirrored volume is striped, be sure to set the policy of the volume to prefer for the striped plex. (The default read policy, select, does this automatically.)

If more than one plex of a mirrored volume is striped, make sure the stripe unit size is the same for each striped plex.

Where possible, distribute the subdisks of a striped volume across drives connected to different controllers and buses.

Avoid the use of controllers that do not support overlapped seeks (these are rare).

The vxassist command automatically applies and enforces many of these rules when it allocates space for striped plexes in a volume.

Do not place subdisks from different plexes of a mirrored volume on the same physical disk. This action compromises the availability benefits of mirroring and degrades performance. Use of the vxassist command precludes this from happening.

To provide optimum performance improvements through the use of mirroring, at least 70 percent of the physical I/O operations should be read operations. A higher percentage of read operations results in a higher benefit of performance. Mirroring may not provide a performance increase or result in performance decrease in a write-intensive workload environment.

	NOTE: The UNIX operating system implements a file system cache. Read requests can frequently be satisfied from the cache. This can cause the read/write ratio for physical I/O operations through the file system to be biased toward writing (when compared to the read/write ratio at the application level).

Where feasible, use disks attached to different controllers when mirroring or striping. Most disk controllers support overlapped seeks that allow seeks to begin on two disks at once. Do not configure two plexes of the same volume on disks attached to a controller that does not support overlapped seeks. This is important for older controllers or SCSI disks that do not cache on the drive. It is less important for many newer SCSI disks and controllers used in most modern workstations and server machines. Mirroring across controllers can be of benefit because the system can survive a controller failure. The other controller can continue to provide data from the other mirror.

A plex can exhibit superior performance due to being striped or concatenated across multiple disks, or because it is located on a much faster device. The read policy can then be set to prefer the "faster" plex. By default, a volume with one striped plex is configured with preferred reading of the striped plex.

For Dirty Region Logging to be in effect, the volume must be mirrored.

At least one log subdisk must exist on the volume for DRL to work. However, only one log subdisk can exist per plex.

The subdisk that is used as the log subdisk should not contain necessary data.

It is possible to "mirror" log subdisks by having more than one log subdisk (but only one per plex) in the volume. This ensures that logging can continue, even if a disk failure causes one log subdisk to become inaccessible.

Log subdisks must be configured with two or more sectors (preferably an even number, as the last sector in a log subdisk with an odd number of sectors is not used). The log subdisk size is normally proportional to the volume size. If a volume is less than 2 gigabytes, a log subdisk of 2 sectors is sufficient. The log subdisk size should then increase by 2 sectors for every additional 2 gigabytes of volume size. However, the vxassist command chooses reasonable sizes by default. In general, use of the default log subdisk length provided by the vxassist command is recommended.

The log subdisk should not be placed on a heavily-used disk, if possible.

Persistent (non-volatile) storage disks must be used for log subdisks.

Make sure that there are enough disks available for the striped and mirrored configuration. At least two disks are required for the striped plex and one or more additional disks are needed for the mirror.

Never place subdisks from one plex on the same physical disk as subdisks from the other plex. Follow the striping guidelines described in “Striping Guidelines”.

Follow the mirroring guidelines described in “Mirroring Guidelines”.

Make sure that there are enough disks available for the striped and mirrored configuration. At least two disks are required for the striped plex and one or more additional disks are needed for the mirror.

Never place subdisks from one plex on the same physical disk as subdisks from the other plex. Follow the striping guidelines described in “Striping Guidelines”.

Follow the mirroring guidelines described in “Mirroring Guidelines”.

Only one RAID-5 plex can exist per RAID-5 volume (but there can be multiple log plexes).

The RAID-5 plex must be derived from at least two subdisks on two or more physical disks. If any log plexes exist, they must belong to disks other than those used for the RAID-5 plex.

RAID-5 logs can be mirrored and striped.

If the volume length is not explicitly specified, it is set to the length of any RAID-5 plex associated with the volume; otherwise, it is set to zero. If the volume length is set explicitly, it must be a multiple of the stripe unit size of the associated RAID-5 plex, if any.

If the log length is not explicitly specified, it is set to the length of the smallest RAID-5 log plex that is associated, if any. If no RAID-5 log plexes are associated, it is set to zero.

Sparse RAID-5 log plexes are not valid.